New robotics and physical-AI papers, with what each one means for the industry.
Reliable real-time failure detection is critical for deploying learned robotic policies, yet defining failures explicitly is infeasible in open-world settings. This motivates failure detection from nominal demonstrations alone.
Scaling robot manipulation policies requires large amounts of diverse, action-labeled data, yet collecting such data in the real world remains expensive and difficult to scale. Existing methods either depend on human teleoperation, rely on simulation with limited visual realism, or use human videos that lack embodiment-specific annotations.
Automating biological laboratories requires robots to perform dexterous tool interactions under realistic visual conditions while handling dynamic experimental phenomena, presenting significant challenges for robotic manipulation. To enable systematic evaluation under these conditions, we introduce AutoBio+, a simulation benchmark and framework for dexterous manipulation in realistic biological laboratory environments.
World-action models (WAMs) predict a robot's actions together with the observations those actions will produce. However, the strongest models rely on video diffusion transformers that are far too computationally expensive for resource-constrained edge devices.
Existing real-world reinforcement learning methods improve robot policies by combining human corrections with autonomous rollouts. However, accumulated experience varies in its value for policy improvement, and even successful trajectories can contain inefficient or redundant behavior, motivating more selective experience reuse.
This paper presents a dual-quadrotor hook- suspended transport system for cooperative aerial manipulation of a payload that is not rigidly attached to a fixed point along the cable. Two quadrotors are connected by a single inextensible cable that passes through a frictionless hook mounted on the payload, allowing the cable to slide between its two segments.
The NIST Assembly Task Board #1 remains a challenging benchmark for learning-based methods in robotic manipulation. Over short horizons, it demands precise, contact-rich control of small, tight-fitting parts.
Autoregressive robot policies rely on action tokenizers to convert continuous action chunks into compact sequences of discrete prediction targets. Existing tokenizers can reconstruct action chunks from discrete codes, but often leave implicit what each individual token contributes to the reconstructed action.
Collecting dexterous manipulation demonstrations for each robot hand requires repeated task-specific data acquisition. We ask whether one shared corpus of natural human demonstrations can instead serve as a reusable data interface for different dexterous hands, without collecting task demonstrations on each target robot.
Action-conditioned world models offer a promising foundation for generalist robotic intelligence by predicting the consequences of actions. However, existing robotic datasets provide limited coverage of the diverse interaction dynamics required for generalizable world modeling.
World Action Models (WAMs) inherit powerful priors from pretrained video generation models, but their variational autoencoder (VAE) representations are optimized for visual reconstruction rather than manipulation. External visual representations can improve manipulation, but existing methods couple representation choices with different integration mechanisms, obscuring their individual effects.
Preprint. The authors present two ways to connect slow frontier reasoning models to fast local robot policies: autonomous generation of corrective demonstrations for training, and dense language supervision at multiple granularities. They report that autonomous demonstration generation becomes more reliable and cheaper as successful examples accumulate, and that a deployment harness lets the frontier model guide the local policy. This matters because it could enable powerful but high-latency models to drive real-time robot manipulation.
This preprint presents AdaTempo, a self-supervised method that speeds up visuomotor imitation policies by extracting shared phase-relative tempo patterns from demonstrations and resampling trajectories before training. It achieves up to a 3.57x speedup over demonstrations, and in real-robot tests raises aggregate success from 58.3% with ACT to 77.8% while completing tasks 2.02x faster. The approach matters because it embeds faster execution directly into standard policies such as ACT or Diffusion Policy without runtime retiming or extra speed-selection modules.
The paper presents Proprioceptive Action Models (PAM), a preprint that adds a compact, timing-free B-spline sketch of the robot’s remaining joint-space path to short-horizon action generation inside one diffusion transformer. Across real-world dual-arm tasks, PAM-DP raises success from 47.5% to 75.0% and also improves Push-T and LIBERO-Long. This offers long-horizon intent without costly language plans or video forecasting.
PEWAM is a preprint describing a 71.5M-parameter world-action model that compiles robot demonstrations into editable “effect programs”—3D object keypoint trajectories, grasp contact points, and final scene configuration—rather than actor-specific motions. On held-out LIBERO-Goal tasks, one program from a single demonstration completes 40 of 90 episodes, while the same backbone under goal-image or language conditioning and prior demonstration-conditioned methods complete at most 19. This matters because turning demonstrations into coordinate-based programs enables editing and cross-robot-arm reuse, a potential route to more flexible robot programming.
DynaHarness is an arXiv preprint that adds a dynamic physical harness around a frozen vision-language-action policy: a slow Qwen3-VL-4B planner proposes capabilities, while a fast 2 Hz governor grounds, monitors, substitutes and recovers them. On LIBERO-Pro it reaches 75.2% success on 800 newly sampled states versus 17.5% for the frozen π0.5 policy, and its self-evolution gains transfer to new states. This gives robot-agent builders a retraining-free way to govern and improve existing policies from execution evidence.
Dream4ACT is a preprint video-action world model that turns target robot joint configurations into four fixed virtual-camera 'action views' rendered from URDF models, so a single model can control different robot arms without per-robot action heads. On RoboTwin 2.0 it averages 88.98% success, scores 65.66 on TriWorldBench, and on two common real-world tasks averages 85.0–90.0% across four robot platforms. It matters because the shared visual action interface removes physical-camera calibration for action encoding and uses training-free joint recovery, simplifying multi-embodiment manipulation models.
This preprint introduces RoboCoach, a world-model-guided coaching loop that decides which subtask demonstrations to collect and which reusable skill expert adapters to update. On real robots, 150 added subtask demonstrations lift complete-task success from 13.3% to 75.0% on Franka and from 40.0% to 83.8% on AgileX, and coached skills transfer to four unseen compositions where a baseline scores 0%. The work matters because it shows world models can direct scarce real-world supervision toward the specific reusable skills that fail, rather than requiring expensive end-to-end data.
World-Action Models (WAMs) couple visual dynamics with action prediction, bringing the rich priors of pretrained video models to robotic manipulation. However, their multi-view interfaces typically tile images or concatenate tokens, leaving the geometric relationships among synchronized cameras implicit.
Robot policies are frequently trained from human corrections, yet teleoperating a robot to provide corrections is burdensome, and human demonstrators are not always optimal. We propose Blended DAgger (BlenDAgger), an approach for collecting data to train imitation learning policies by using shared control to blend the policy's and demonstrator's actions during interventions.
Papers come from arXiv robotics feeds. Where our summary is not written yet, you see the opening lines of the abstract.