Early-Signal Detector

architecture · first seen 13 Oct 2026

LLM agent for robotics

Also: agentic robotics

Placing a large language model planner in the robot control loop to compose perception and motion tools.

Agentic robotics puts a large language model agent inside the robot control loop, where it calls perception, motion and vision-language-action policy tools, or writes and refines robot code, instead of training a dedicated policy per task. Recent papers move from proposing the idea to engineering it. One cuts per-step LLM reasoning cost by chunking actions. Another schedules when the LLM reviews a VLA policy. Others contribute a benchmark for industrial variational automation and a controlled study of 28K coding-agent rollouts. A Grid Robotics position paper frames it as a new paradigm, and a curated reading list and a Google DeepMind skills plugin for simulation and real robots appeared on GitHub.

If it holds up, robot capability would come from general foundation models plus tool libraries rather than per-task training, which shifts the open problems to inference cost, supervision scheduling and benchmarks for reliability.

Analyst note (emerging), 18 Nov 2026.

Mentions per week

0219 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Rank 10 on 16 Nov 2026 · 6 mentions in 28 days · 5 independent groups · papers → code · first seen 5 weeks ago · acceleration +1.10

Earliest sources

Latest sources

Mentioned together with

action chunkingrobotic manipulation benchmarkembodied AI evaluationMuJoCo