New robotics and physical-AI papers, with what each one means for the industry.
Reinforcement Learning (RL) substantially improves planning performance in autonomous-driving Vision-Language-Action (VLA) models. However, existing RL methods rely on scalar trajectory rewards that measure overall quality but fail to identify specific planning errors or provide actionable correction signals.
Vision-language-action (VLA) models and world action models (WAMs) have advanced robotic policy learning by building on pretrained vision-language models (VLMs) and video generative models (VGMs), respectively. Yet VLAs learn without explicit supervision from future world states, while WAMs that condition actions on predicted futures incur extra inference cost and let visual prediction errors propagate into actions.
Vision-language-action (VLA) policies issue sparse targets whose arrival at the robot controller is delayed and irregular, so a reasonable predicted action can still produce abrupt motion near people. Remedies are evaluated on different streams, delays and metrics.
Pre-trained robot policies, such as Vision-Language-Action (VLA) models, provide a strong foundation for generalist robotic manipulation, but adapting them to new tasks with only a few demonstrations remains challenging. Directly fine-tuning a pre-trained policy often achieves strong in-distribution performance, yet it can distort the representations learned during pre-training and weaken the model's generalization ability under out-of-distribution shifts.
Pretrained vision-language-action (VLA) models encode broad manipulation priors, yet adapting them to specific tasks requires balancing limited robot interaction, computational resources, and safety constraints. Noise space learning (NSL) adapts flow-matching-based VLAs by learning a context-conditioned distribution over noise while keeping the pretrained VLA fixed.
Actions specify what a robot should execute, but do not describe what those commands will achieve. Vision-language-action (VLA) policies nevertheless learn primarily by predicting action chunks from images, language, and state.
A user who brings a generalist manipulation policy into their home or office wants it to pick up their objects from their shelves, counters, and tables. Yet even a strong generalist like π_(0.5) achieves 0% pickup success in our captured deployment scenes without adaptation, and collecting real demonstrations in every user's environment is prohibitively expensive.
Scaling generalist robotic policies is limited by the high cost of collecting embodied interaction data. Large-scale egocentric human videos offer a rich alternative, but they lack robot action labels, and existing action annotation methods are often noisy under real-world visual complexity and diverse manipulation dynamics.
A preprint from Chinese humanoid maker Galbot and the Chinese Academy of Sciences tests which off-the-shelf image encoders make video-based robot brains more reliable, and finds that one strong general-purpose encoder delivers most of the gain. Their resulting recipe tops the RoboCasa humanoid tabletop benchmark at 84.7% and adds 10 points of success on three real Galbot robot tasks.
SimpleTouch is an arXiv preprint that adds a tactile expert to the π0.5 vision-language-action model, using a frozen pretrained T3 tactile encoder and training only on 50 task demonstrations with future-touch prediction as auxiliary supervision. It reaches 77.5% average success across six UniVTAC contact-rich tasks and 71.3% across four real-robot tasks, outperforming the pretrained tactile policy FTP-1 by 10.8 and 8.8 percentage points. This suggests strong contact-rich manipulation can be learned without costly tactile policy pretraining or explicit visuotactile alignment.
This preprint introduces FastOPD, a method for distilling large vision-language-action (VLA) and world-action-model teachers into compact few-step robot policies. It keeps 84% of a 3.3B-parameter teacher's LIBERO success rate with only two sampling steps, reduces inference latency by 78.1%, and on RoboTwin 2.0 lifts single-step SmolVLA success from 35.3% to 51.2%. The work matters because it may allow foundation-policy capabilities to run on resource-constrained, low-latency robot hardware.
This preprint introduces ManiPhysicsBench, a simulation-based benchmark that scores vision-language-action manipulation policies on object preservation, not just task completion. It combines a curated 91-object material property library with post-rollout physics-solver damage checks in LIBERO and SimplerEnv; public VLAs average 13.5% safe success versus 48.3% task success on one platform. The work matters because real-world deployment of fragile-object handling depends on not damaging goods, yet current benchmarks ignore this risk.
This preprint introduces MixVLA, a model-agnostic training strategy for vision-language-action (VLA) policies that separates invariant and non-invariant latent representations, then adaptively mixes the non-invariant branch to regularize environment-specific variation. On LIBERO-Plus it posts a 76.2% average zero-shot success rate versus 69.6% for OpenVLA-OFT, and on RoboTwin it lifts π0.5 clean-to-randomized success from 46.0% to 62.4%. Because it needs no extra out-of-distribution data or architecture changes, it is a practical robustness upgrade for existing VLA stacks.
This preprint introduces SARI, a phase-split co-training framework that post-trains a vision-language-action policy on simulated free-space approaches and real contact interactions for contact-rich manipulation. On a lamp task, it reduces real demonstration time by 34.3% and reaches 27.5% success at unseen object placements, where full-task simulation, real-only, and full-task co-training baselines score 0%. The result shows how to allocate scarce real contact data while preserving spatial generalization.
This preprint introduces MobiAgent, a dual-loop framework for long-horizon mobile manipulation. It decouples high-level VLM planning from low-level flow-matching skills and adds an automated lifelong learning loop from deployment data. The abstract reports no quantitative results, but the approach matters because it targets compounding errors and interference between locomotion and arm control.
Preprint. The authors build an inference-time defense for vision-language-action (VLA) policies against adversarial image patches by training a sparse autoencoder, identifying an internal feature that tracks patch presence, and suppressing that feature only when a linear probe detects an attack. On LIBERO-10 under intermittent UADA attacks, conditional suppression raises success rate, e.g. from 26.0% to 32.4% for π0.5 at n=10, while avoiding the large benign-condition drop caused by continuous suppression. This points to lightweight, fine-tuning-free robustness layers for deployed robot policies.
In this preprint, the authors introduce DuoMind, a distributed multi-robot framework that pairs per-robot VLM orchestrators with low-level VLA action policies and structured semantic messages. Across two simulation benchmarks, it raises long-horizon success substantially over finetuned single policies, for example 39.25% vs 1.00% on Cook Pot and 78.00% vs 52.25% on Hang Bag. The work matters because it shows multi-robot coordination can be added to off-the-shelf foundation models without centralized control or joint multi-agent training.
This arXiv preprint describes ChunkVLA-AM, a deployment pipeline that adapts an OpenVLA-OFT vision-language-action model to a FAIRINO FR3 robot for additive-manufacturing post-print retrieval. The robot predicts and executes 8-step chunks of 7-D Cartesian actions in a cloud-edge loop and completed 39 of 42 physical pick-and-transfer trials, a 92.9% success rate. The work provides a concrete feasibility check for using VLA-based automation in a fixed manufacturing workcell.
This preprint tunes action-expert vision-language-action models to accept non-language task cues: 8-channel EMG envelopes or visual segmentation annotations. On a real cube-picking task, EMG conditioning lifts cluttered success from 30% to 50% and cuts failure rate from 50% to 4%; on 16 simulated RoboCasa pick-and-place tasks, visual annotations lift cluttered success from 11% to 20% and cut failure from 47% to 25%. The findings suggest downstream VLA policies can usefully condition on continuous cues beyond fixed language prompts, especially in clutter.
UniWAM is a preprint introducing a unified robot foundation model that combines a vision-language reasoner, a video-generation world model, and an action predictor. It reaches 99.2% success on LIBERO, 92.6% on LIBERO-Plus OOD perturbations, 68.32% on RoboTwin 2.0 Clean2Random, and 82.5% real-robot instruction-following rate. The work matters because it shows how co-training on robot, human egocentric, and VQA data can improve robustness, generalization, and long-horizon manipulation.
Papers come from arXiv robotics feeds. Where our summary is not written yet, you see the opening lines of the abstract.