Agent-as-Policy2026cc-by-4.0

YAM Agent-as-Policy — Real Dual-Arm LLM-Agent Manipulation Trials

196 real-robot sessions where an LLM agent acts directly as the policy on a bimanual YAM arm, reading camera observations and issuing Cartesian and joint commands across nine manipulation tasks. Every session includes synchronized camera streams, depth frames, 50 Hz joint trajectories, full agent event streams, and human-graded outcomes.

Downloads4
Episodes196
Likes1

Why This Matters for Physical AI

This dataset demonstrates large language models operating as end-to-end policies for real-world bimanual manipulation, providing insights into LLM reasoning, tool use, and long-horizon task planning with real robot feedback and human evaluation.

Technical Profile

Modalities
rgbdepthproprioceptionlanguage
Robot Embodiments
YAM dual-arm
Action Space
joint_positions, end_effector_delta
Environment
lab
Task Types
manipulationblock_stackingdie_flippingtowel_foldingpart_insertionassemblytool_use
Episodes
196
Data Format
parquet, WebDataset, PNG
Annotation Types
language_instructionsreward_labelsaction_labelshuman_grading
License
cc-by-4.0
Part of the Agent-as-Policy family

Access

Need custom rgb data?

Claru builds purpose-built datasets for lab applications with dense human annotations and quality assurance.

Request a Sample Pack

Related Datasets