Agent-as-Policy2026cc-by-4.0

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

162 real-robot trials where an LLM agent acts directly as the policy on a bimanual YAM arm setup, reading camera observations and issuing Cartesian and joint commands across ten manipulation tasks.

Downloads30
Episodes162
Likes2

Why This Matters for Physical AI

This dataset provides a comprehensive record of LLM agents directly controlling dual-arm manipulation in real-world settings, enabling research into how language models can serve as policies for long-horizon robotic tasks with full step-by-step reasoning transparency.

Technical Profile

Modalities
rgbdepthproprioceptionlanguage
Robot Embodiments
YAM bimanual arm
Action Space
joint_positions, end_effector_delta
Environment
lab
Task Types
manipulationblock_stackingdie_flippingtowel_foldingpart_insertionassemblythrowingtool_use
Episodes
162
Data Format
parquet, WebDataset
Annotation Types
language_instructionsreward_labelsaction_labels
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