ChunkAct: Action Chunking for Agentic Robotics
arXiv cs.AI · 14 Oct 2026, 03:59 AEST
ChunkAct introduces action chunking to agentic robotics, allowing for more efficient execution of tasks. This method improves task success while significantly reducing the number of LLM calls and input/output tokens required.
Key facts
- Agentic robotics places a large language model (LLM) agent in the robot control loop, where it composes perception, motion, and vision-language-action (VLA) tools under the guidance of a memory and thereby generalizes to new tasks without retraining.
- Every primitive requires a round of LLM reasoning over the full interaction history, so the cost of an episode grows with the number of primitives.
- On LIBERO-Pro, ChunkAct improves task success by 5.0 points over the state of the art in agentic robotics while requiring 2.9x fewer LLM calls, 4.4x fewer input tokens, and 2.3x fewer output tokens.
Read aloud
ChunkAct introduces action chunking to agentic robotics, allowing for more efficient execution of tasks. This method improves task success while significantly reducing the number of L L M calls and input/output tokens required. Agentic robotics places a large language model L L M agent in the robot control loop, where it composes perception, motion, and vision-language-action VLA tools under the guidance of a memory and thereby generalizes to new tasks without retraining. Every primitive requires a round of L L M reasoning over the full interaction history, so the cost of an episode grows with the number of primitives. On LIBERO-Pro, ChunkAct improves task success by five point zero points over the state of the art in agentic robotics while requiring two.9x fewer L L M calls, four.4x fewer input tokens, and two.3x fewer output tokens.
Sources in this cluster
- ChunkAct: Action Chunking for Agentic Robotics
arXiv cs.AI · 14 Oct 2026, 03:59 AEST