For practitioners and technical leaders
An engineering mapof Physical AI.
Awesome Physical AI is a curated, engineering-oriented catalog across 14 categories — simulators, datasets, benchmarks, evaluation methodology, foundation models, world models, manipulation, locomotion, sim-to-real, safety, governance, production patterns, courses, and companies. Researchers and ML/robotics engineers use it to shortlist tooling and reference work; technical leaders use it to scope evaluation, risk, and deployment for embodied AI systems. Selection lens over hype.
Browse by category
14 categories, each with a selection lens.
Every category page opens with what it is, why it matters for evaluation or deployment, and how to choose between entries — followed by a short, vetted list.
Simulators
Physics engines and rendering stacks where policies are trained and stress-tested before touching hardware.
Datasets
Recorded robot behaviour — teleoperation, demonstrations, egocentric video — that feeds imitation learning and pretraining.
Benchmarks
Fixed task suites for comparing policies under controlled conditions across manipulation, locomotion, and embodied reasoning.
Evaluation Methodology
How to measure policies correctly — harnesses, metrics, and statistical practice that turn rollouts into defensible claims.
Robotics Foundation Models
Generalist pretrained policies and vision-language-action (VLA) models that map perception and language to robot actions.
World Models
Learned predictive models of physical dynamics used for imagination-based planning, sample-efficient RL, and synthetic data.
Manipulation
Grasping, dexterous, and contact-rich methods — analytic planners, diffusion and transformer policies, and the rigs that feed them.
Locomotion
Legged, bipedal, and humanoid controllers and the GPU-parallel training stacks that make modern sim-to-real practical.
Sim-to-Real
Techniques for closing the reality gap — domain randomisation, system identification, residual learning, and real-robot fine-tuning.
Safety & Robustness
Constrained training, adversarial evaluation, and failure-mode analysis that decide whether a system is shippable.
Governance & Policy
Standards, regulation, and risk frameworks that shape how Physical AI systems are evaluated and deployed.
Production Patterns
Middleware, planners, fleet orchestration, and observability for shipping robots in production.
Courses
Structured learning paths into robot learning and embodied AI for individuals and team onboarding.
Companies
Organisations actively shaping Physical AI in industry — foundation-model labs, platform builders, and applied deployers.
Next steps
Decide whether this fits your work.
Read the overview
Mission, audience, and how the catalog is organised — the fastest way to decide whether this resource fits your work.
Curation standards
The selection bar every entry must clear: maintained, relevant, distinctive, and load-bearing for real engineering work.
Scope & limits
What this list does not claim. Useful for technical leaders evaluating where the catalog ends and your own diligence begins.