Robotics & Physical AI

UNITAS is a 3D-native world action model for embodied manipulation. It represents observations, robot or human motion, and scene changes in a shared metric 3D frame.

GroundingPI is a 4B-parameter model designed to make precise visual grounding a native capability for embodied AI. It reaches 73.68% on average across 34 grounding benchmarks and improves downstream robotics and driving systems.

ENCORE turns a small number of robot demonstrations into structured evidence that a coding agent can inspect and use to write manipulation policies. The approach improves performance on LIBERO-PRO, solves otherwise underspecified RoboDojo tasks, and transfers to a real bimanual robot.

Artisan combines a multi-agent generation pipeline with physics-simulation feedback for articulated 3D assets. Instead of optimizing appearance alone, it targets objects that can move, interact, and function in simulation.

EmbodiedCLBench brings continual-learning evaluation into embodied simulation, testing whether agents can reuse basic skills in unseen task compositions. Its experiments show that experience helps, but simply adding more samples does not produce proportional gains.

Amazon’s warehouses now rely on more than one million robots, yet the company still employs over 1.5 million people. The next phase of automation is less about removing humans entirely and more about changing how they work alongside machines.

TouchScale is a 500-hour visual-tactile dataset collected with a unified wearable setup. Its synchronized egocentric video and dense bimanual touch signals improve cross-sensor contact prediction and real-world robot manipulation in the reported experiments.

The Reflex framework trains a humanoid to infer the motion of a thrown box from egocentric RGB-D observations and coordinate its entire body to catch it. It reaches 85.7% success in simulation and demonstrates real-world catching on a Unitree G1.

SuperNav introduces an agentic navigation system for service robots that lets a pretrained multimodal model interpret requests and scenes while specialized navigation tools handle motion execution. The approach targets generalization across tasks and unfamiliar environments.

SimpleICL revisits a basic question in robot in-context learning: what should a robot actually learn from a visual demonstration? The work combines a clearer task definition with a minimal visual-prompt encoder and a low-cost data collection recipe.

UniWAM presents a unified world-action model that combines physical reasoning, visual world generation, and action prediction in one training framework. It brings together human egocentric videos, robot demonstrations, and visual question answering data to improve embodied understanding and control.

NVIDIA researchers introduce Long-WAM, a model-and-system framework that brings longer visual history to real-time robot control. Its key insight is that autoregressive video pretraining matters more than simply enlarging the context window.

RobotWorld is a simulation benchmark for testing whether multimodal agents can turn instructions, visual observations, and tool use into reliable robot behavior. Its results show that agents can build sophisticated perception and control pipelines, but still struggle to maintain state, recover from failure, and verify completion.

DiVeR identifies decision-critical states in robot trajectories by measuring the dispersion of sampled action representations. It uses that signal to focus verifier training where action selection has the greatest effect on task success.

A robot may execute a command differently from the motion predicted by a vision-language-action policy because of friction, backlash, payload changes, or wear. A new approach adapts the VLA during deployment by learning from this command-to-motion gap, without task labels or rewards.

EyeRobot 2.0 actively steers two eye viewpoints toward 3D fixation targets, concentrating visual computation where manipulation matters most. The approach explores precise bimanual control with a single stereo camera and no wrist-mounted cameras.

MotorMind is a robotics harness that connects a general-purpose vision-language model to deterministic robot control through mid-level actions and continuous feedback. The approach explores whether a VLM can perform zero-shot manipulation without task-specific policies, coding agents, or extra grounding tools.

Skill2Real introduces an agentic framework for transferring robot manipulation skills through a shared API rather than task-specific policy retraining. Its hierarchical memories and Proposer–Verifier–Governor loop deliver strong results on simulated benchmarks and frozen real-world skills.

PyRUA-Lean replaces repeated tool calls with executable Python cells that compose robot primitives, perform local checks and request only necessary observations. Across 700 simulated instances, it raised success from 63.1% to 71.7% under the same LLM-call budget while cutting input tokens by 65% on jointly solved tasks.

A remembered object location can become wrong while a robot is on its way to inspect it. EvolvingNav combines time-indexed 3D memory, probabilistic belief updates, and visual evidence to support navigation in evolving environments.

A new survey organizes robot in-context learning into four execution interfaces and examines how demonstrations can preserve task requirements as environments change. It also proposes a clearer way to evaluate teaching responsiveness, physical transfer, and experience reuse.