Robots¶
| Robot | Benchmark | Runs on | VLA policy | Cameras the agent sees |
|---|---|---|---|---|
robodojo_xpl_rgb |
RoboDojo | XPolicyLab, under RoboDojo's own evaluation client | none (SAM3 for segmentation) | three colour cameras |
libero (default) |
LIBERO-PRO | RPent | Pi0.5 (and SAM3) | agentview, wrist |
robotwin |
RoboTwin 2.0 | RPent | LingBot-VLA | head, two wrists |
robocasa |
RoboCasa365 | RPent | RLDX-1 | agentview, navview, wrist |
New robots are added through XPolicyLab (Adding a Robot): the benchmark's own client runs the simulator and judges the episode, and one adapter directory can serve the benchmarks XPolicyLab integrates (today: RoboDojo). LIBERO-PRO, RoboTwin 2.0 and RoboCasa365 run on RPent's stacks.
Every robot page gives the API reference the agent reads, generated from the robot class when these docs are built.