Physical AI terms with source trails.

AI systems that reason about, simulate, or act in the physical world through robots, vehicles, sensors, and control loops.

Models built to generate, reason about, or predict physical scenes and trajectories for training, testing, or planning.

Models that connect visual observations and language instructions to robot actions.

Reasoning that accounts for spatial layout, physical affordances, task state, and what an embodied agent can do next.

Robot trajectories, observations, actions, labels, task descriptions, and environment metadata used to train or evaluate policies.

Training data collected from multiple robot bodies, sensors, action spaces, and task environments.

Generated or simulated observations, trajectories, scenes, and annotations used to supplement real robot data.

An open-source robotics library and ecosystem focused on robot learning datasets, policies, training, and hardware deployment workflows.

Software environments that approximate physics, sensors, robots, tasks, and environments for training or validation.

A computational engine for dynamics, contact, constraints, actuation, and other physical interactions.

Structured digital replicas of robots, spaces, assets, or processes used for simulation and operational reasoning.

The transfer of policies, perception, or validation results from simulated environments to real hardware.

Combining signals from cameras, depth sensors, force sensors, proprioception, IMUs, lidar, or tactile arrays.

Estimating the current state of the robot, objects, environment, and task from noisy observations.

Contact-sensitive sensing used for manipulation, grasping, slip detection, and material interaction.

Physical interaction with objects: grasping, pushing, folding, opening, placing, tool use, and deformable-object handling.

Moving the robot body through the world: wheels, legs, balance, gait, terrain, and recovery.

Robots with human-like body plans designed to operate in environments shaped for humans.

The mapping from robot observations and goals to actions, often learned from demonstrations, reinforcement learning, or model outputs.

Planning feasible paths, trajectories, or action sequences under geometry, dynamics, and task constraints.

The lower-level control, timing, safety interlocks, middleware, and execution layers beneath high-level policies.

Coordinating many joints and contacts so the whole robot body acts coherently.

Common robotics middleware used to connect sensors, actuators, messages, nodes, tools, and robot software components.

The compute inside or near the robot that runs perception, policy, planning, safety, and communication workloads.

The delay between observation, model computation, command generation, and hardware actuation.

The power, battery, heat, and cooling constraints that bound onboard model size and runtime.

Structured tests used to compare models, policies, simulation quality, task success, generalization, or safety-relevant behavior.

Testing failure modes, unsafe actions, out-of-distribution scenes, human interaction, and recovery behavior.