Get Started
Home
Topics
Search
Library
Research questionHow can LLMs synthesize executable machines and improve their designs through physical simulation feedback?Conventional program-synthesis evaluations often test symbolic or digital outputs rather than spatially assembled mechanisms. This leaves unclear whether LLM agents can create machines that satisfy physical tasks and use simulation feedback to make useful structural revisions.
AI
AI Agents
Code Generation & Program Synthesis
Evaluation & Benchmarks
Reinforcement Learning
Latest papersRecent research connected to this question, newest first.Compositional Machine Design as Program Synthesis with LLMsThe evidence concerns BesiegeField, a Besiege-based simulation in which agents compose machines from standardized parts, execute them, and receive rewards and state feedback. It covers single-agent generation, iterative editing, hierarchical workflows, reinforcement learning from simulation-derived rewards, and human reference performance; it does not provide evidence from physical hardware.research paper · Sep 1, 2026
Related questions
How can LLM orchestrators preserve continuous state when collaborating with non-language agents?How can we train and evaluate LLM agents for tool use across single- and multi-turn workflows with serial or parallel calls?How can LLM agents reuse execution traces without losing temporal and outcome-dependent behavior?How can LLMs reason reliably about physical systems from large, fine-grained simulator traces?