Daily Papers Arch&EAI

Vision-language-action (VLA) models commonly adopt an LLM-centric $V \to L \to A$ pathway, where visual observations are projected into the representation space of a large language model before being decoded into robot actions. Although effective, this design incurs substantial computation and memory overhead at every policy invocation. In this work, we introduce TurboVLA, a new VLA paradigm that reformulates the conventional $V \to L \to A$ pathway as a direct $V + L \to A$ mapping. Instead of using a large language model as the central interface between perception and action, TurboVLA independently encodes visual observations and language instructions, directly exchanges information between them through lightweight bidirectional vision-language interaction, and predicts continuous action chunks with a compact decoder. This simple design constructs task-conditioned representations directly from visual and linguistic features, significantly reducing the computational and memory costs of VLA inference. On LIBERO, TurboVLA achieves 97.7% average success with only 0.2B parameters, 31.2 ms inference latency, and 0.9 GB inference VRAM on a consumer-grade RTX 4090, matching or outperforming substantially larger VLA policies. These results establish TurboVLA as a simple and effective alternative to the prevailing LLM-centric VLA paradigm, offering a new perspective on how vision, language, and action can be connected for efficient robotic manipulation. Code is available at https://github.com/H-EmbodVis/TurboVLA.

Vision-language-action (VLA) models commonly adopt an LLM-centric $V \to L \to A$ pathway, where visual observations are projected into the representation space of a large language model before being decoded into robot actions.

Vision-language-action (VLA) models remain constrained by scarce action-labeled robot data, whereas action-free videos offer abundant observations of physical change. Latent action models can extract such priors, but reconstruction-trained codes may predict future observations without the structure required for joint generation with robot actions. Existing structured methods add temporal constraints but retain deterministic transition points, so residual errors in locally inferred transitions may propagate and compound under recursive composition. We introduce DLAM, a distributional latent-action model that represents each transition as a diagonal Gaussian. Reconstruction conditioned on the reference frame grounds the mean in observed visual change, while normalized composition and reversal over equal-gap triplets constrain both the mean and dimension-wise variance. Variance composition uses a lightweight shared-correlation coefficient to account for dependence between adjacent transitions that share an intermediate frame, whereas reversal negates the mean and preserves the variance. For downstream policy learning, we freeze the encoder and train a flow-matching policy to jointly generate mean transition sequences and robot actions. On held-out transitions, DLAM learns more temporally consistent latent dynamics than existing latent-action baselines and achieves stronger direct and cumulative reconstruction on held-out videos. Under the same controlled $π_0$ transfer protocol, it also improves policy performance on MetaWorld MT50, LIBERO, and real-world manipulation tasks. Controlled ablations show that normalized mean constraints account for most of the reconstruction gain, while learned variance and correlation-aware composition provide complementary improvements in downstream control.

Vision-language-action (VLA) models remain constrained by scarce action-labeled robot data, whereas action-free videos offer abundant observations of physical change.

We introduce VEGA, an approach for training navigation VisionLanguage-Action (VLA) models from unlabeled egocentric navigation videos. Internet-scale egocentric videos provide a scalable source of navigation-relevant visual observations, capturing cluttered scenes, close-range obstacles, and natural human motion through real-world spaces. However, these videos are not directly usable for policy learning because they do not provide obstacle-aware trajectories conditioned on explicit navigation goals in the robot's coordinate frame. VEGA addresses this gap by reconstructing local scene geometry from monocular video, sampling navigation goals (represented as text, image, or spatial waypoints) and generating obstacle-aware trajectories using the constructed geometry. The resulting trajectory distribution is then used to train a flow-matching VLA navigation policy. By using geometry exclusively during training, VEGA distills obstacle-aware planning directly into a vision-based policy. Furthermore, we introduce VEGA-Bench, a benchmark containing 250k scenes and approximately 5 million navigation goals paired with scene geometry, designed to evaluate goal progress, collision avoidance, and obstacle clearance of VLAs. Our evaluation shows that VEGA achieves competitive goal progress while reducing collisions by 33.0% and improving obstacle clearance by 17.9% over the strongest baseline on VEGABench, while improving success by at least 150.0%, reducing collisions by at least 66.7%, and improving obstacle clearance by at least 60.0% in real-world trials. Ultimately, we demonstrate that video-derived geometric supervision provides a scalable and effective signal for training obstacle-aware navigation VLAs. The code and benchmark will be released at the time of publication.

We introduce VEGA, an approach for training navigation VisionLanguage-Action (VLA) models from unlabeled egocentric navigation videos.

Despite the impressive visuomotor capabilities enabled by Vision-Language-Action (VLA) models, their performance often degrades on challenging and out-of-domain tasks. Recent test-time steering and scaling methods improve performance without extensive data collection and retraining, but action samples often remain concentrated around similar behaviors and therefore inherit correlated failure modes. Moreover, existing methods apply the same intervention strategy at every timestep, regardless of whether the base policy is already likely to succeed. To address these limitations, we introduce $RL^2$, an adaptive inference-time steering framework that leverages Reinforcement Learning on VLA Latents. First, we train a lightweight offline RL policy conditioned on expressive latents extracted from the VLA action expert and compose its flow velocity with that of the frozen VLA during inference. This compositional steering strategy combines the behavioral priors of large-scale imitation learning with the action diversity induced by offline RL beyond dominant demonstration modes. We further discover that inference-time steering follows fundamentally different scaling laws under success and failure states, revealing that action diversity is most beneficial when the base VLA is likely to fail, but can unnecessarily perturb already-accurate actions when success is likely. Building on this insight, $RL^2$ activates compositional steering only when failure is predicted. Across the SIMPLER and PolaRiS benchmarks, $RL^2$ improves success rates by up to +17.3% in out-of-domain settings, while ablations and scaling studies demonstrate the importance of latent representations and RL training. Finally, real-world experiments demonstrate that these gains transfer beyond simulation, establishing $RL^2$ as a practical and modular steering framework for VLA deployment.

Despite the impressive visuomotor capabilities enabled by Vision-Language-Action (VLA) models, their performance often degrades on challenging and out-of-domain tasks.

Deep reinforcement policy learning directly in physical robots (on-robot learning) remains bottlenecked by slow wall-clock training times. We present SymmGrid, a trajectory level augmentation framework inspired by parallelized symmetries that super-scales group transformations to significantly accelerate on-robot learning in both egocentric and exocentric visual setups. We model a Markov Decision Process (MDP) under a symmetry tree, in which state-action pairs have admissible parallelized invariant transformations that yield a geometric grid structure. The state is modelled with ego- or exocentric images and proprioception information. The latter require special treatment, in the form of homographies, to warp visual scenes in line with their corresponding spatial transformations. These parallelized transformations produce a large set of unique symmetric equivalences that populate the replay buffer with diverse and consistent experiences that speed up learning and improve performance. We present extensive training and evaluations performed directly on real robot manipulation contact tasks including peg-insertions, cable routing, and object relocations. Relative to SOTA, SymmGrid achieved wall-clock training convergence speed-ups of 1.37-2.17x, evaluation success rate improvements of 1.09x-1.27x, fastest training convergence times of 16.6, 10.9, and 79.3 minutes respectively. For trajectory wide assessments, we used normalized area under the curve (nAUC) ratios. SymmGrid achieved improvements of up to 2.59x. These results confirm that simple branch symmetries can have an outsized result due to super-scaling and bring us closer to sub-10 minute on-robot learning training in manipulation tasks suitable for arms and humanoids. The project page is available at symmgrid-robot.github.io

Deep reinforcement policy learning directly in physical robots (on-robot learning) remains bottlenecked by slow wall-clock training times.

Biomedical laboratory robots must navigate to instruments before performing experimental procedures. Existing embodied navigation platforms are designed for household environments and treat a target as an object center or an arbitrary nearby position. This representation is inadequate for laboratory instruments, which must be approached from their operating side while maintaining safe clearance from surrounding equipment. We introduce BioVLN, a simulation platform for developing and evaluating visual-language navigation agents in biomedical laboratories. BioVLN represents each instrument with three regions: its physical body, a surrounding clearance region, and an operation area in front of the usable side. This model is applied consistently to scene generation, target placement, navigation evaluation, and safety analysis, so success depends on reaching a position from which the instrument can be accessed. BioVLN supports procedural scene generation and manually designed environments, producing 47 scenes and 1667 episodes. Standardized navigation and reinforcement-learning interfaces enable trajectory collection and policy training. Experiments show that geometric exploration reaches 74.4--87.5% success, while sampling multiple valid positions in the operation area improves success to 83.3--92.5% and reduces unsafe proximity.

Biomedical laboratory robots must navigate to instruments before performing experimental procedures.

The key bottleneck in embodied AI is not model architecture but data. Although billions of human manipulation videos exist online, robots cannot directly learn from them due to the embodiment gap between human morphology and robot hardware. We introduce Pegasus, a low-resource framework that bridges this gap by translating human demonstrations into robot-learnable data through structured knowledge transfer. Instead of relying on raw video prompts, Pegasus constructs a graph-based intermediate representation: a Task Graph extracted from human videos is transformed through Affordance and Constraint Graphs into a Robot Planning Graph for robot-conditioned video generation. A hierarchical affordance latent space models the relationship between object states, affordances, and tasks, enabling generalization beyond object identities. A closed-loop physics verifier further filters invalid generations using kinematic feasibility, collision constraints, and joint limits. We evaluate Pegasus across a range of egocentric manipulation benchmarks, including GTEA Gaze+ and EPIC-KITCHENS-100, and diverse robot embodiments, assessing Task Correctness, Executability, State Consistency, and Learnability. Results demonstrate reliable cross-embodiment translation and show that robot data generation can be reframed from a hardware collection problem into a scalable, low-resource knowledge transfer problem.

The key bottleneck in embodied AI is not model architecture but data.

General-purpose robotic manipulation requires robots to perform diverse tasks in open-world environments while improving their skills over time. Despite recent progress in robotic manipulation, existing systems still primarily acquire manipulation skills in a static manner, where capabilities are learned for specific tasks or settings rather than adaptively evolving through physical interaction. Resembling how repeated practice enables humans to develop muscle memory, advanced manipulation proficiency requires an autonomous capability evolution mechanism that allows robots to progressively transform interaction experiences into increasingly effective manipulation abilities. To this end, we propose HERO, a self-improving hierarchical embodied agent that enables autonomous capability evolution from zero human demonstrations. HERO organizes heuristic reasoning, exemplar reuse, and reflexive execution into a unified orchestration framework, allowing robots to autonomously bootstrap manipulation experience, rapidly accumulate reusable behaviors through experience transfer, and progressively consolidate recurring interactions into efficient closed-loop visuomotor policies. By tightly coupling autonomous data collection with task execution, HERO continuously expands and dynamically schedules manipulation capabilities according to different stages of experience accumulation and execution requirements. Extensive experiments demonstrate that HERO substantially reduces human intervention during robotic data collection while achieving robust manipulation across diverse tasks, providing a promising path toward self-improving robotic systems.

General-purpose robotic manipulation requires robots to perform diverse tasks in open-world environments while improving their skills over time.

While MoE augments VLA via expert specialization, router suffers from ineffective expert routing owing to the kinematic heterogeneity of actions across manipulation tasks and, even worse, the unavailability of the kinematic signals at inference time. In this work, we first observe that most semantically distinct manipulation tasks reduce to multiple kinematic archetypes. Motivated by this finding, we propose Kinematics-supervised explicit routing (KinRT), a new paradigm that shifts from implicit, observation-driven expert routing to explicit, kinematics-guided expert dispatching. Specifically, we perform kinematic clustering on action trajectories into multiple kinematically coherent groups, whose IDs serve as ground truth to supervise the training of the router; at inference time, the router dispatches experts only using visual-language observations, without any reliance on action kinematics. KinRT actually introduces an asymmetric bridging mechanism that distills the task kinematics from the action space in training into the observation space at inference. In addition, to assess KinRT's cross-platform generalization, we build an economical, Do-It-Yourself robot (DIYRobot) platform from scratch using 3D-print technology ($<$ 2,000USD). Extensive experiments demonstrate KinRT's superiority over both dense and MoE-featured VLAs by more than 23.26% on RoboTwin benchmark and 20.27% on our introduced DIYRobot platform. Our code and DIYRobot platform will be open-sourced.

While MoE augments VLA via expert specialization, router suffers from ineffective expert routing owing to the kinematic heterogeneity of actions across manipulation tasks and, even worse, the unavailability of the kinematic signals at inference time.

Vision-language-action (VLA) policies commonly execute long-horizon mobile manipulation through open-loop action chunks, issuing multiple actions without receiving new high-level visual input. A committed chunk therefore implies how observations should evolve, but accidental deviations can violate this expectation while the remaining actions continue to propagate the error: commit-time policy confidence cannot react to a deviation that occurs after dispatch, and observation-only anomaly scores lack an action-conditioned reference for separating expected effects from unexplained changes. We propose CheckVLA, which verifies execution with a separately trained, frozen action-conditioned world model. A conformally calibrated risk threshold bounds the episode-level probability of an unnecessary first intervention and determines when to intervene, its exceedance controls how strongly the rewritten suffix retains the superseded chunk, latency-aware hard prefixing restricts replacement to actions that remain deployable, and an event-driven keyframe bank preserves evidence of prior progress across repairs. On RoboCasa365, under a common training recipe and a matched invocation budget, CheckVLA attains a 36.1% average success rate against 27.6% for periodic replanning (+8.5 points). At a matched 5% episode-level false-alarm target, action conditioning raises timely recall to 77.9%, against 48.6% for an observation-only control and 37.9% for an action-shuffled control. These simulation results support action-conditioned verification as a way to restore feedback during chunked execution while keeping the repair consistent with inference latency.

Vision-language-action (VLA) policies commonly execute long-horizon mobile manipulation through open-loop action chunks, issuing multiple actions without receiving new high-level visual input.

Wearable cardiovascular sensor patches promise continuous, unobtrusive monitoring, but their tight energy, memory, and compute budgets make it unclear whether physiological signals should be analyzed on the device or streamed to the cloud for processing. We study this inference-versus-transmission trade-off for a resource-constrained patch that records synchronized electrocardiogram (ECG) and phonocardiogram (PCG) signals. We propose an end-to-end, multi-modal convolutional neural network (CNN) with early fusion that classifies the two modalities directly on the device, without hand-crafted features. Trained and validated on the PhysioNet/Computing in Cardiology Challenge 2016 dataset, the floating-point model attains an accuracy of 0.975, which is competitive with the best reported results. At the same time, it reduces the parameter count and computational cost by approximately three orders of magnitude. We deploy an 8-bit integer version of the model on a microcontroller with an integrated neural processing unit (NPU) and measure its inference energy. We also benchmark the energy required for Bluetooth Low Energy (BLE) communication on a representative evaluation kit across a range of payload sizes. NPU inference consumes approximately one-seventh of the energy required for CPU inference. For realistic per-second payloads, local inference is also several times more energy efficient than continuous raw-data streaming. These results show that on-device intelligence, rather than constant transmission, is the more energy-efficient basis for always-on wearable cardiovascular monitoring at the edge.

Wearable cardiovascular sensor patches promise continuous, unobtrusive monitoring, but their tight energy, memory, and compute budgets make it unclear whether physiological signals should be analyzed on the device or streamed to the cloud for processing.

Tiny Vision-Language-Action models are appealing for real-time robotic control, but reducing model scale often weakens two capabilities essential for manipulation: task-conditioned spatial grounding and coherent action generation. We introduce XS-VLA, a lightweight framework that teaches tiny VLA policies "where to look" and "how to move" without increasing deployment-time model cost. For spatial grounding, Coarse-Grained Spatial Distillation uses Qwen3-VL-4B to produce teacher-derived coarse image-plane location labels for task-relevant objects, which are quantized into a spatial vocabulary and distilled into a SmolVLM2-0.25B backbone without human annotations. For action generation, Latent Flow Matching combines a CVAE-style latent variable with flow-based policy learning to organize multimodal demonstrations during training and produce stable action chunks at deployment. On LIBERO, XS-VLA improves average success from 82.8\% to 90.3\% over SmolVLA-0.25B and improves LIBERO-Long from 63.0\% to 89.0\%. In real-world experiments, we design three tasks covering single-arm placement, precision bimanual stacking, and long-horizon sequential coordination, where XS-VLA improves average task success from 21.7\% to 65.0\%. These results show that explicit spatial grounding and latent action-structure learning can make tiny VLA models effective for robotic manipulation.

Tiny Vision-Language-Action models are appealing for real-time robotic control, but reducing model scale often weakens two capabilities essential for manipulation: task-conditioned spatial grounding and coherent action generation.

Current Vision-Language-Action (VLA) models rely mainly on 2D inputs, neglecting the rich object structural information and commonsense knowledge inherent in the 3D physical world. This deficiency restricts their spatial awareness and adaptability for complex, high-precision manipulation. To bridge this crucial gap, we construct a Concept Expert module for VLA to build executable Analytic Concepts that represent objects as explicit, programmatic blueprints. Our mechanism operates in two synergistic phases: First, prior to VLA inference, the Concept Expert leverages 3D information from Vision Foundation Models (VFMs) to estimate the initial kinematic and structural parameters. Second, throughout the manipulation process, the VLA model utilizes its inherent capability to dynamically track the dynamic concept parameters, continuously aligning them with observational changes to ensure persistent accuracy. Once established, the Analytic Concepts provide explicit, high-quality guidance for VLA fine-tuning through (1) dense, programmatic manipulation rewards and (2) precise spatial guidance. This formulation allows VLA models to learn physically grounded interaction behaviors while maintaining end-to-end learning flexibility. Our experimental results show consistent improvements in success rate and learning efficiency across supervised and reinforcement learning settings, demonstrating the effectiveness of structured, concept-based guidance for VLA post-training.

Current Vision-Language-Action (VLA) models rely mainly on 2D inputs, neglecting the rich object structural information and commonsense knowledge inherent in the 3D physical world.

The pursuit of high-performance PDE solvers rests on three fundamental challenges: (i) the curse of dimensionality in kinetic and financial equations, (ii) the poor extrapolation of purely data-driven surrogates, and (iii) the widening gap between algorithm design and hardware specialization. To overcome these challenges, we present NeuroPDE+, a scalable neuromorphic PDE solver design based on spintronic and ferroelectric devices for accelerating PDE solutions. NeuroPDE+ consists of two dedicated units: a diffusion tracking unit (DTU), which emulates random walks on Markov chains through activations between hardware neurons, and a scattering tracking unit (STU), which samples non-local jumps via a multi-level probability tree. System-level simulations suggest that NeuroPDE+ achieves a squared error below 1e-2 in steady-state heat equation and particle transport problems. Simulation results further indicate that the DTU achieves up to a 315x performance gain over previous neuromorphic processors, and that the STU achieves a 1000x speedup compared to a general-purpose CPU. Co-designing algorithm and hardware with intrinsic stochasticity and non-volatile in-memory computing, NeuroPDE+ preliminarily explores a new paradigm for efficient and scalable neuromorphic PDE solvers. This approach could pave the way for probabilistic computing architectures in large-scale scientific simulations.

The pursuit of high-performance PDE solvers rests on three fundamental challenges: (i) the curse of dimensionality in kinetic and financial equations, (ii) the poor extrapolation of purely data-driven surrogates, and (iii) the widening gap between algorithm design and hardware specialization.

To operate effectively across diverse contexts, robots must not only perform manipulation tasks accurately but also adapt how their actions unfold to the task, object, and interaction setting. We ask whether this execution-level variation can be learned as a reusable behavioral factor shared across tasks. We present \textbf{MoMo}, a two-stage imitation-learning framework consisting of a spatiotemporal action tokenizer and a behavior-cloning transformer that takes task and a continuous motion-mode condition as inputs. Across six real-robot manipulation tasks, varying this condition produces steady, dynamic, and intermediate behaviors that human raters can distinguish and that differ in joint speed, acceleration, and end-effector approach pitch. On tasks demonstrated in only one mode, MoMo transfers the unseen requested mode while largely preserving task success. Together, these results provide evidence of compositional generalization to unseen task--mode combinations and show that motion mode can be reused across tasks to control how a manipulation skill is performed.

To operate effectively across diverse contexts, robots must not only perform manipulation tasks accurately but also adapt how their actions unfold to the task, object, and interaction setting.

We present TiPToP, a modular manipulation system that integrates pretrained foundation models with a GPU-accelerated Task and Motion Planner to solve tasks directly from RGB images and natural language. TiPToP composes perception, planning, and execution modules and requires no robot training data. It can be deployed on a standard DROID setup in under an hour and adapted to new embodiments with minimal effort. We evaluate TiPToP against $π_{0.5}\text{-DROID}$, a state-of-the-art VLA fine-tuned on 350 hours of demonstrations, across two real-world DROID setups (one operated by an external team) and simulation, where TiPToP attains a higher average success rate and faster average completion time. We also evaluate on the MolmoSpaces benchmark, where TiPToP ranks first overall on pick and pick-and-place tasks among methods not trained on in-distribution data. We further show that TiPToP's modularity enables us to trace failures to specific components, revealing where to target improvements. We release TiPToP open-source to serve as a reproducible baseline and to enable further research on modular manipulation systems. Project website and code: https://tiptop-robot.github.io

We present TiPToP, a modular manipulation system that integrates pretrained foundation models with a GPU-accelerated Task and Motion Planner to solve tasks directly from RGB images and natural language.

LLMs are widely regarded as alien intelligences, systems whose cognitive operations are fundamentally unlike our own. Apparent similarities to human cognition are therefore often seen as the result of anthropomorphic projection. We argue that this framing is mistaken. LLMs clearly differ from humans in important respects, including their physical substrate, learning history, and the environments with which they interact. These differences make it all the more striking that contemporary LLM-based systems converge with human cognition on a number of principles of cognitive organization with longstanding support in cognitive science. We identify structural correspondences across five dimensions: inferential organization, computational architecture, representational structure, prediction-driven learning, and reinforcement-learning-like mechanisms supporting goal-directed action. These correspondences support a broader model of intelligent cognition in which core principles long used to explain human intelligence also characterize contemporary LLM-based systems.

LLMs are widely regarded as alien intelligences, systems whose cognitive operations are fundamentally unlike our own.

Acoustic information provides rich cues about object location, material properties, and changes caused by contact or motion. This paper introduces a new set of acoustic-aware manipulation tasks for imitation learning, in which robots must use auditory cues to determine manipulation targets. These tasks require sound source localization and identification for active exploration in robotic manipulation. Also, we propose a multimodal imitation learning framework, Spatial-Spectral Audio Action (S2A2), that integrates visual features with acoustic spatial and acoustic signal information for the acoustic-aware manipulation tasks. We implemented S2A2 models that integrates policies such as ACT, Diffusion Policy, VQ-BeT, and $π_0$, into our framework. Simulation experiments showed that the proposed method is the most effective for tasks requiring both position and timbre. Furthermore, real-robot experiments confirm the applicability of the proposed tasks and framework to real-world manipulation.

Acoustic information provides rich cues about object location, material properties, and changes caused by contact or motion.

Embodied intelligence integrates learned perception and decision making with real-time computation, control, and physical interaction. Because failures can cause immediate physical or operational harm, task completion alone does not establish trustworthiness. We define trustworthy embodied intelligence as the sustained capacity to execute specified tasks reliably under environmental and system variation while maintaining risk within acceptable bounds. We term this objective sustained safe success. Its supporting mechanisms are organized into four interdependent layers. The model layer generates task-competent action proposals with calibrated uncertainty and explicit safety preferences. The system layer realizes authorized actions dependably through integrated sensing, computation, control, hardware safeguards, fault containment, and fallback. The evidence layer substantiates bounded claims through evaluation, verification, validation, traceability, and structured assurance arguments. The deployment layer maintains claim validity through runtime monitoring, authority management, intervention, incident response, and controlled updates. Because assumptions and failures propagate across these layers, neither model capability, isolated safeguards, nor benchmark performance alone can establish end-to-end trustworthiness. Drawing on embodied AI, robotics, control, dependable computing, distributed systems, and autonomous driving, we further propose a non-normative hierarchy of trustworthiness levels. This hierarchy grades the strength of bounded deployment claims across task capability, safety, system assurance, operational governance, and supporting evidence, providing a basis for bounded deployment, comparative evaluation, research prioritization, and future standardization.

Embodied intelligence integrates learned perception and decision making with real-time computation, control, and physical interaction.

Current work on robot failure detection and correction typically operates in a post hoc manner, analyzing errors and applying corrections only after failures occur. This work introduces CycleVLA, a system that equips Vision-Language-Action models (VLAs) with proactive self-correction, the capability to anticipate incipient failures and recover before they fully manifest during execution. CycleVLA achieves this by integrating a progress-aware VLA that flags critical subtask transition points where failures most frequently occur, a VLM-based failure predictor and planner that triggers subtask backtracking upon predicted failure, and a test-time scaling strategy based on Minimum Bayes Risk (MBR) decoding to improve retry success after backtracking. Extensive experiments on the LIBERO and LIBERO-Plus simulation benchmarks show that CycleVLA surpasses the state-of-the-art VLA π0.5, improves success rates by correcting execution failures across VLAs of varying capability, from under-trained ones to fully converged policies, and that MBR serves as an effective zero-shot test-time scaling strategy for VLAs. On a real robot, CycleVLA reaches a 91% average success rate on one precise and two long-horizon manipulation tasks. We further conduct stress tests with multiple manually injected perturbations (e.g., swapping in a distractor at the expected location while relocating the true target object mid-execution), where CycleVLA corrects ~80% of injected failures and maintains success rates comparable to unperturbed execution. Project Page: https://dannymcy.github.io/cyclevla/

Current work on robot failure detection and correction typically operates in a post hoc manner, analyzing errors and applying corrections only after failures occur.

Recently, photonic transformer accelerators (PTAs) have successfully achieved significant speedup and energy efficiency improvements over electronic accelerators for expediting Transformer inference. However, state-of-the-art rely on expensive multi-wavelength light generation and large dot-product units due to active phase-shifter components, thus making their approach inefficient and impractical. To address this, we propose MDTransformer, a novel hardware-software co-design of PTA based on mode-division optical dataflow and operations. Specifically, MDTransformer performs complex matrix operations using spatial-mode interference, that leverages the inverse-designed multi-mode couplers, crossings, and Mach-Zehnder IQ modulators into a compact mode-division photonic tensor core (MPTC), capable of executing matrix multiplications in the optical domain. Its each guided mode (i.e., TE0-TE3) acts as an independent computational lane, enabling four-fold parallelism-per-waveguide without spectral filtering or free-spectral-range limitations. Moreover, its coherent detection and IQ modulation jointly encode amplitude and phase, realizing complex-valued arithmetic for full-range operations in transformers. MDTransformer offers analog multiplication with sub-4-bit effective precision and inter-modal crosstalk below -30 dB. Its inverse-designed approach also offers scalable and full compatibility with single-laser continuous-wave operation at 1550 nm. Experimental results show that MDTransformer achieves 40.4% area reduction, 63.6% power saving, 40.6% energy saving, and comparable latency over the state-of-the-art PTA across different workloads (i.e., DeiT-Tiny/Small/Base and BERT-Base/Large). These results show that MDTransformer offers a practical solution for high-performance and energy-efficient transformer-based systems.

Recently, photonic transformer accelerators (PTAs) have successfully achieved significant speedup and energy efficiency improvements over electronic accelerators for expediting Transformer inference.

Vision-Language-Action (VLA) models have shown strong potential for general robot manipulation, but most existing models rely on 2D visual-language backbones and lack fine-grained 3D understanding of target objects, especially under occlusion, pose variation, scale changes, and precise spatial interaction. We propose an object-centric 3D representation alignment framework built upon $π_0$, using SAM3D as a frozen 3D teacher to provide target-object 3D priors during training. Specifically, we localize task-relevant objects with object recognition models, generate corresponding object masks, and use SAM3D to extract dense object-level 3D representations, which are aligned with intermediate visual features of $π_0$. This enables the policy to internalize target-object 3D information while preserving the original RGB-language-to-action inference pipeline without requiring depth, point clouds, masks, SAM3D, or additional 3D modules at test time. Simulation experiments show consistent improvements, achieving 99.1\% on LIBERO and an average length of 4.11 on CALVIN. Real-world experiments further demonstrate that our method is particularly effective in long-horizon manipulation scenarios where the robot must focus on different target objects across multiple subtasks.

Vision-Language-Action (VLA) models have shown strong potential for general robot manipulation, but most existing models rely on 2D visual-language backbones and lack fine-grained 3D understanding of target objects, especially under occlusion, pose variation, scale changes, and precise spatial interaction.

Generating robust and reactive manipulation strategies that can adapt to changing context information is a challenging task in robotics. Over the years, Learning from Demonstration (LfD) has emerged as an intuitive and effective solution for generating reactive policies, particularly by following dynamical-system(DS)-based approaches. However, most state-of-the-art DS-based approaches focus on addressing the robustness limitations, overlooking the modulation of policies in response to the environment. As a result, they tend to be inflexible with respect to parameterization by task-dependent variables. In this work, we build on existing work on policy fusion and uncertainty quantification to propose a context-adaptive policy framework that combines task-parameterized, robust and reactive manipulation. For this, we use LfD to acquire a policy that is conditioned on the robot state and low-dimensional task-dependent parameters reflecting the environment. We combine the learned policy with additional uncertainty-aware policies using a Mixture of Experts (MoE) formulation to improve its out-of-distribution (OOD) robustness and convergence behavior. The approach is evaluated on the LASA handwriting dataset and on a real 7-DoF robot in three scenarios: force-conditioned grasping, manipulation of deformable food items and object-centric grasping.

Generating robust and reactive manipulation strategies that can adapt to changing context information is a challenging task in robotics.

Pre-trained Vision-Language-Action (VLA) models provide useful semantic and spatial priors, yet their parallel-gripper action interfaces do not specify how those priors should be realized by a dexterous hand. Directly appending finger joints conflates two decisions with different structure: when contact should be established and how a morphology-specific hand trajectory should establish it. We introduce InDex, an intent-conditioned adaptation framework that separates these decisions without discarding full hand supervision. InDex derives a normalized grasp intent from retargeted demonstrations. A first stage predicts synchronized end-effector--intent chunks; conditioned on these predictions, VLA context, and proprioception, a diffusion decoder generates multi-joint hand actions. The scalar intent is therefore a temporal coordination interface rather than a compressed hand pose. Across four simulated tasks, three VLA backbones, and a physical arm--hand platform, InDex preserves the VLA's reaching competence while markedly improving conversion from approach to stable grasp and task completion. Ablations isolate complementary roles: intent aligns the contact transition, whereas diffusion represents the multiple hand trajectories compatible with the same task-space plan. These results identify post-reach arm--hand coordination, rather than object localization alone, as the principal bottleneck in adapting parallel-gripper VLAs to dexterous manipulation.

Pre-trained Vision-Language-Action (VLA) models provide useful semantic and spatial priors, yet their parallel-gripper action interfaces do not specify how those priors should be realized by a dexterous hand.

An often overlooked factor of robot manipulation performance is the embodiment of the robot itself. Motivated by this problem, we study motion-conditioned robot co-design, where the goal is to generate complete robot designs that track target end-effector trajectories (from human demonstrations) while optimizing user-defined rewards. We introduce Transformer Transformer, a diffusion transformer trained on RoboTokens, a unified tokenization of robot embodiments, states, and actions. The same architecture can be used across embodiment spaces (e.g., wheeled bimanual, quadrupeds, humanoids) and use cases (embodiment generation, cross embodiment controller). Rather than overfitting to one reward function, Transformer Transformer is a dynamics model, whose reward-agnostic state and action predictions can be converted into reward-specific value predictions. These value predictions are used to steer embodiment diffusion towards high value robot designs, through a procedure we call Dynamics Self-Guidance. Experiments across multiple design spaces show zero-shot optimization of unseen rewards and trajectories, improving performance and runtime over the evolutionary baseline. Finally, we fabricated an optimized ALOHA design, which reduced tracking error by over 70% compared to the original design.

An often overlooked factor of robot manipulation performance is the embodiment of the robot itself.

Vision-Language-Action (VLA) models provide a promising foundation for general-purpose robotics, yet their real-world deployment demands the ability to continually acquire new skills without forgetting prior ones. While recent studies have explored continual learning for VLA models in simulated settings, the challenge remains largely unexamined under realistic physical conditions. To bridge this gap, we construct a real-world continual learning benchmark comprising ten diverse sequential manipulation tasks across both single-arm and bimanual configurations. Through extensive experiments on this benchmark, we find that naive sequential fine-tuning leads to severe catastrophic forgetting, whereas a well-configured experience replay (ER) approach can effectively mitigate forgetting and outperform joint multi-task training under equivalent computational budgets. Notably, by synthesizing our empirical findings, we successfully achieve stable continual learning across the full 10-task heterogeneous stream, retaining previously acquired capabilities while adapting to diverse new skills in real-world deployment. This work presents an empirical study grounded in real-world continual VLA learning and offers actionable insights for deploying robust, long-lived robotic policies.

Vision-Language-Action (VLA) models provide a promising foundation for general-purpose robotics, yet their real-world deployment demands the ability to continually acquire new skills without forgetting prior ones.

Robotic hardware evolves over time, but demonstration data is often tied to a specific sensor and actuator configuration. This raises a practical and underexplored question: when does legacy data begin to benefit an upgraded robot? We study this question on a wheeled humanoid platform across two hardware generations, where both the camera and gripper are changed while the overall morphology remains fixed. Contrary to the common assumption that more cross-configuration data is always helpful, we observe a grokking-like transition: legacy data remains ineffective until the upgraded configuration acquires a minimum level of task competence, after which co-training gains rise sharply before diminishing near saturation. We hypothesize that this task-dependent transition is governed by a transfer threshold and characterize the resulting three-phase pattern. Across real-robot manipulation tasks, we observe all three phases: no measurable benefit at low competence ($10.0\% \rightarrow 10.0\%$), a sharp gain after crossing the threshold ($23.3\% \rightarrow 86.7\%$ on flower insertion), and diminishing returns at high competence ($85.0\% \rightarrow 93.3\%$ on pen insertion). We provide a theoretical account based on gradient alignment and residual policy uncertainty, and derive a phase-aware rule for deciding when to collect more new-hardware data and when to reuse legacy demonstrations. We further validate this three-phase pattern on a mobile dual-arm watering task, with results consistent with our predictions.

Robotic hardware evolves over time, but demonstration data is often tied to a specific sensor and actuator configuration.

Deploying diabetic retinopathy (DR) screening models in primary care requires edge-efficient systems that remain accurate, safe, and reliable under domain shift. Multi-teacher knowledge distillation (KD) is a natural compression strategy, but existing approaches largely assume that all teachers provide equally trustworthy supervision. In our setting, this assumption fails: a strong CNN teacher (EfficientNet-B3, 0.876 QWK) and a weaker Transformer teacher (Swin-Base, 0.830 QWK) are complementary, yet the Transformer's logits can still mislead the student. We therefore propose OrthKD, a selective-trust distillation framework that transfers full supervision from the strong CNN, uses feature-only distillation from the weak ViT, and enforces orthogonality between teacher-specific student projections to encourage complementary rather than redundant evidence. This design preserves local lesion precision, injects global structural context, and improves robustness to distribution shift. On 132,049 retinal images, a 5.4M-parameter MobileNetV3 student reaches 0.885 QWK on EyePACS and improves zero-shot Messidor-2 performance from 0.507 to 0.728 QWK, while also achieving strong referral AUC and calibration. These results show that selectively distilling heterogeneous teachers can enable practical DR screening on resource-constrained devices.

Deploying diabetic retinopathy (DR) screening models in primary care requires edge-efficient systems that remain accurate, safe, and reliable under domain shift.

Vision-language-action (VLA) models predict sequential actions to execute tasks specified by language instructions, conditioned on visual observations and proprioceptive states. However, how to fuse modalities in VLA models remains an open problem, since robot manipulation involves dynamic phases, such as long-distance movements and close-range interactions, in which the importance of visual observations may vary over time. In this paper, we propose an infer-diagnose-refine (IDR) framework, a model-agnostic framework that can be integrated with diverse VLA architectures for refining action predictions at test time. IDR first infers actions under factual and counterfactual scenarios of visual observations, and then diagnoses the causal effects of visual observations as the estimated dynamic importance, which is finally used to refine the action predictions in a training-free manner. We further design a causality-aware action refiner to realize the IDR framework, including zero-padding interventions for inferring counterfactual actions, norm-based quantification for diagnosing causal effects, and gated residual fusion for refining actions. Extensive experiments on both simulation benchmarks and real-world tasks show improvements in overall performance across multiple VLA backbones, demonstrating the efficacy of dynamically adjusting visual importance at test time.

Vision-language-action (VLA) models predict sequential actions to execute tasks specified by language instructions, conditioned on visual observations and proprioceptive states.

Vision-Language-Action (VLA) models translate natural-language commands into robot action sequences, but leading systems on the LIBERO-Plus robustness benchmark use three- to seven-billion-parameter backbones whose memory demands can exceed embedded robotic budgets. We present CoTinyVLA, a 0.9B-parameter action model on a Qwen3.5-0.8B backbone that obtains that robustness by structuring supervision instead of enlarging the model. Three components target different axes of the problem: dual-view temporal input of 16 history frames per step with textual camera and time markers; hierarchical chain-of-thought (CoT) distillation from a 35B teacher into an episode-level Plan and a chunk-level Think span over task phase, gripper state and next subaction; and paraphrase augmentation expanding 40 base commands into 800 variants. On LIBERO-Plus, spanning 10,030 perturbed tasks across seven perturbation dimensions, CoTinyVLA reaches 90.8% on Spatial, 87.3% on Object, 86.6% on Goal and 80.7% on Long, leading the strongest 7B baseline on all four suites by 4.7, 2.8, 15.9 and 3.0 points, with every margin interval excluding zero. The gains concentrate on the hardest axes of the benchmark: across the eleven published baselines none exceeds 53.2% on Robot Initial States in any suite, whereas CoTinyVLA reaches 73.6% on Goal against 39.9% for the strongest baseline. Ablations show the three components to be separable by perturbation axis, and at a matched image budget how frames are divided between the two cameras and across time accounts for 8.6 points on its own. Closed-loop inference peaks at 2.25 GiB of allocated GPU memory, and paired interventions show the episode Plan to be load-bearing: replacing it with an empty or contradictory span costs 40 to 45 points of success. Structured supervision thus lets a 0.9B backbone exceed all of them. Code: https://github.com/BrainJellyPie/CoTinyVLA

Vision-Language-Action (VLA) models translate natural-language commands into robot action sequences, but leading systems on the LIBERO-Plus robustness benchmark use three- to seven-billion-parameter backbones whose memory demands can exceed embedded robotic budgets.