The Light CompanyRobot evaluation resources.
Guides, methods and examples for teams running physical robot evaluations.
Evaluation guides
How to plan a real-world robot evaluation
Plan a robot evaluation with defined tasks, controlled conditions, scoring rules and a record of every trial.
Read guide →Why robot evaluation starts with the reset
Object placement, camera setup and scoring rules can change an evaluation. Learn how to separate intended conditions from what actually happened.
Read guide →Capability maps for robot evaluation
Go beyond a single robot success rate. Organize evaluation by position, orientation, clutter, objects and lighting while preserving untested conditions.
Read guide →Scene history for robot evaluation
How scene versions connect physical conditions, policy checkpoints and run evidence for repeatable robot evaluation.
Read guide →
Prism in the lab
See projected staging and real workcells in the lab gallery. Watch an evaluation from setup to review, or follow the product workflow.
The technical brief explains initial-condition control and the physical constraints involved.
Research references
Robot Learning as an Empirical Science: Best Practices for Policy Evaluation discusses reporting initial conditions, success criteria and observed behavior.
Colosseum studies robustness across environmental changes. SIMPLER studies simulated evaluation against real robot setups.
Product walkthroughs
Explore scene history with an interactive example, or open the Physical AI Index to compare models across sixteen Prism benchmarks.
For machine-readable product information, see llms.txt and the public agent guide.