New robotics and physical-AI papers, with what each one means for the industry.
Vision-language-action (VLA) models connect visual perception and language understanding with robot control, providing a foundation for instruction-conditioned manipulation. However, conventional pretrained VLA policies lack a mechanism for gradual adaptation through accumulated interaction experience during deployment.
Executing a vision-language-action (VLA) policy involves deciding when to continue an attempt and when to intervene to improve its chances of success. We introduce ACRO, an Actor--Critic rollout orchestration layer that makes these decisions for a fixed VLA.
A Stanford and Princeton preprint shows a robot policy that drops the wrist camera and instead learns where to zoom in on a fixed external camera image, lifting real-world task success to 70.8% with one camera and 77.1% with two, against 52.1% for the same base model with a wrist camera. Fewer cameras on the arm means cheaper, sturdier robots and training data that could come from human videos recorded without wrist views.
Foundation models are driving a paradigm shift in robotic intelligence, moving beyond low-level action prediction to goal-directed orchestration of data collection, model training, task planning, and real-world execution. We characterize this emerging paradigm as an agentic robots, in which agents reason over human instructions, interact with robotic environments, rectify execution errors autonomously, and dynamically adjust operational strategies.
Sim-and-real co-training is an effective way to scale robot learning by augmenting limited real-world demonstrations with simulation data. However, existing approaches typically treat simulators as fixed data sources, relying on uniform generation, offline retrieval, or manual parameter alignment by domain experts.
Robotic manipulation is inherently geometric, but what shape does the world take inside the policy that performs it? Prior work on vision-language-action (VLA) policies establishes that task-relevant variables are linearly readable from activations, yet readability certifies only that a variable is present and says nothing about the shape its values trace, which may be curved and multi-dimensional.
A single-author preprint from start-up Dxtr AI turns a human demonstration filmed on a phone into a record of what the object did, then checks in simulation which of eight robot bodies can reproduce it: 31.2% of 157 robot-episode pairs passed, 52.2% failed and 14.0% were refused up front. It is a data-filtering tool, not a trained policy, but it targets the expensive step where each robot needs its own task data.
Vision-language-action (VLA) models have shown strong potential for general-purpose robotic manipulation, but extending them to bimanual dexterous manipulation introduces a high-dimensional and structured action space. Existing bimanual VLA policies commonly concatenate right- and left-side arm--hand actions at each control timestep, leaving cross-hand interactions implicit within a single action vector.
Vision-language-action (VLA) models have recently achieved remarkable progress by leveraging large-scale robot datasets and foundation models for general-purpose robot manipulation. However, their ability to rapidly adapt to novel manipulation tasks beyond the training distribution remains limited.
Vision-Language-Action (VLA) models are gaining attention in autonomous driving, inheriting commonsense priors from large language models (LLMs) and widening the navigation interface from a fixed command set to free-form natural-language instructions. However, this open-ended interface exposes a new vulnerability surface: an inattentive driver may issue instructions that are unsafe given the current scene, and a malicious user may issue dangerous ones on purpose.
Though modern AI models like LLMs predict discrete tokens, today's robotic foundation models represent actions as continuous vectors. In this paper, we challenge this common consensus by asking if continuous action parametrizations are superior to their discrete counterparts.
Recent vision-language-action (VLA) models have shown strong performance on local robot manipulation, yet their success does not readily translate to long-horizon mobile manipulation. In our experiments, a pretrained subtask VLA achieves 80% success rate on atomic manipulation tasks but only 19% when these skills are composed into mobile manipulation.
Vision-language-action (VLA) policies fine-tuned on LIBERO are trained and evaluated on the same ten instruction strings, and rewording the instruction changes base OpenVLA's success by -38.0 points. Test-time prompt ensembling, which improves language models, does not recover this loss in any setting we ran.
Peking University researchers built a common test bed for the question of how a robot AI model should use a wrist force sensor, and found that the simplest option, writing the force reading into the model's text prompt, worked most consistently. On a real arm it lifted average task success from 40% to 63%, while fancier time-series designs hurt on assembly work.
Progress toward general intelligence in the physical world requires scaling robot learning across diverse environments and interactions. However, existing robot policies use shared parameters that remain fixed across interaction contexts.
A Seoul National University preprint makes robot foundation models act faster on embedded chips without retraining: it samples a few slightly perturbed action plans per call and executes more steps when the plans agree. On NVIDIA Jetson Orin and Thor this cut inference time by up to 2.3 times while success rates stayed level or rose, including 93.5% against 90.7% on a real UR10e arm.
Quadruped robots performing outdoor deliveries must interpret complex scenes, anticipate changes, and navigate in socially compliant ways. Robot demonstrations are costly to collect at the scale needed for generalization.
Surface navigation on the Moon is subject to extreme illumination contrast and deployment-time perceptual disturbances. While modular navigation pipelines have long supported planetary rover navigation, robustness is compromised by measurement errors that propagate through hand-engineered components.
Robotic disassembly of end-of-life (EoL) products requires robots to accommodate product variation, follow prerequisite sequences, and approach components precisely for contact-rich manipulation. Supervised fine-tuning (SFT) provides a practical path for adaptation, but limited and costly disassembly data cannot capture the distribution shifts arising from changing environments and product uncertainties.
Vision-language-action (VLA) policies perform manipulation tasks using camera images and language instructions, but object slip or visual occlusion can disrupt execution. Recovery is typically learned during policy training or delegated to a monitor that triggers a scripted repair after detecting a failure.
Papers come from arXiv robotics feeds. Where our summary is not written yet, you see the opening lines of the abstract.