New robotics and physical-AI papers, with what each one means for the industry.
Translating language into humanoid motion requires coordinating body parts, determining motion duration, and producing references suitable for robot execution. We introduce PartiGen, a part-aware latent diffusion framework that incorporates anatomical structure into a shared whole-body motion representation.
End-to-end planning in complex scenes must model diverse driving intentions and select the most appropriate trajectory using scene context and future risks. However, multi-query regression may collapse to one mode, while anchor-based and generative planners share a scaling bottleneck: small candidate sets limit maneuver coverage or discovery of high-quality trajectories, whereas large sets increase generation and scoring costs.
Robot policies often convert sensing and control into fixed-rate observation-action sequences, leaving physical time implicit in sequence position. This assumption breaks when observation rates or delays vary, or when inference latency changes.
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.
Nearly every autonomous or semi-autonomous flipper controller assumes access to an elevation map of the terrain under and in front of the robot. The terrain beneath the chassis is occluded in the sensors, so the map is remembered through odometry integration, whose drift corrupts it.
Autonomous landing of unmanned aerial vehicle (UAV) on unmanned surface vehicle (USV) is difficult in high sea states. Existing hierarchical landing methods often infer future USV motion only implicitly from short observation histories, which limits their ability to adapt to severe and rapidly changing sea conditions.
Humanoid tennis requires reactive whole-body coordination under rapidly changing ball dynamics, yet collecting task-specific motion-capture demonstrations that cover diverse approach states, contact locations, and timing conditions is costly. Inspired by how humans acquire new skills by combining reusable motor experience with task-specific demonstrations, we present ATHLETE, a framework for learning reactive humanoid tennis from generic locomotion data and ordinary tennis videos.
Generating edge case scenarios is crucial for training, testing and safety assuring autonomous systems. This identifies hazardous failures to increase system reliability and robustness, ensures the system can still function safely on the boundaries of its operational design domain and learns the expected system behaviour for these edge cases.
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.
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.
SafeStreamingFlow is a preprint describing a start–goal-conditioned flow matching planner that streams one-step state updates in physical execution time, so safety filtering only needs to act on the step actually executed. In simulations, the safe version reaches 88% success with 4% violations on Maze2D Large (vs 13% success and 71% violations for SafeDiffuser) and runs Hopper safety filtering with 7.30 ms total latency vs 1012.4 ms for SafeDiffuser. The result matters because it brings execution-time safety filters to generative planners without the usual O(H*K) compute and distribution-shift failures.
This preprint presents a controller that turns empowerment, an information-theoretic measure of future controllability, into an objective for sampling-based model predictive control. It uses the same action distribution for probing and executed control and needs only first-order derivatives. On simulated pendulum swing-up tasks with sparse rewards, combining this intrinsic objective with a task cost improved success rates up to 100%, while task cost alone failed.
This preprint introduces NAVA-WAM, a world action model that directly pretrains a robot action policy from observation-only videos, avoiding separate visual or latent-action interfaces. It reports 83.5% success on LIBERO-Plus, 73.6% on the RoboTwin 2.0 random domain, and 93.3% on real-robot tabletop trials versus 66.7% for DreamZero and 53.3% for π0.5. The approach matters because it offers a route to scale robotics control priors from abundant unlabeled video rather than expensive action annotations.
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 presents DR-IPC, a LiDAR-based quadrotor navigation stack that feeds online disturbance estimates into a nonlinear model-predictive controller for joint obstacle-aware motion and control. In Gazebo multi-goal trials with wind and a suspended payload, DR-IPC completes 9 of 10 missions versus 1 of 10 for the prior IPC baseline, and in an indoor archway experiment it reduces altitude RMSE from 0.34 m to 0.01 m. The system runs onboard at 100 Hz, making disturbance-resilient cluttered flight more practical.
Preprint. RoboBridge is a framework that represents manipulation tasks as editable procedural skills and wraps a frozen vision-language-action policy as a guided action tool, evolving those skills from execution feedback. On LIBERO-PRO it reaches 74.4% overall success, beating the strongest reported baseline at 72.1%; on three real-robot tasks, 10 episodes of adaptation lift success from 9/15, 7/15, 7/15 to 15/15, 14/15, 15/15. This shifts sim-to-real transfer from retraining to selective skill revision, cutting adaptation iterations by 25% and token use by about 75%.
In a preprint, the authors present SceneFactory-3D, a GPU-batched multi-agent driving simulator that uses per-wheel tire forces, suspension, and 3D terrain so road friction and grade physically change how vehicles execute commands. Across 1,024 matched 12-vehicle worlds per learned policy and 21 road conditions, valid work-zone passage dropped by 6 to 90 percentage points when surface friction fell from 1.0 to 0.18, and near-collision time-to-collision events increased for every learned policy. This matters because it shows safety conclusions from traffic simulation are sensitive to road physics that most scalable simulators ignore.
OmniAct3D is a preprint framework that adapts frozen vision-foundation-model 3D detectors to equirectangular 360° images. It improves NDS by 2.96 points over the prior best on Spheriverse and mAP by 24.87 points over the unadapted VFM baseline on PanoMMOcc. This gives mobile robots a single-image continuous 360° 3D detection path while preserving transferable foundation geometry.
Preprint: HexVIO offloads the stereo-inertial VIO frontend of Basalt onto the Hexagon DSP in an unmodified Samsung Galaxy S23. It delivers a 67% whole-device power reduction at the CPU baseline's maximum throughput, or up to 86% higher throughput at matched power, while preserving accuracy. At paced 30 fps it tracks at 0.83 W, equivalent to about 18 hours on the device's 15 Wh battery, which matters for all-day XR, wearables, and robotics.
This preprint presents a self-repairing recurrent ensemble policy that recovers from sensor shifts online without expert labels. The headline result is that random observation masking plus Kalman fusion and per-step RFLO updates lets simulated locomotion policies regain near pre-shift performance after a sudden sensor shift, while full-visibility ensembles collapse and do not recover. The work matters for unattended robots that must self-correct when sensors drift, fail, or become noisy.
Papers come from arXiv robotics feeds. Where our summary is not written yet, you see the opening lines of the abstract.