Reproduce “PLATO: Planning with LLMs and Affordances for Tool Manipulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Arvind Car, Sai Sravan Yarlagadda, Alison Bartsch, Abraham George, Amir Barati Farimani
PLATO, developed at Carnegie Mellon University, is an LLM-centric framework that enables robots to perform complex manipulation tasks, including zero-shot tool use, from natural language instructions without prior environmental knowledge. It achieved up to 70% success in single-task grasping and 60% in single-task tool use, leveraging GPT-4o for planning and an affordance model for task-oriented grasping.
Key insights
- LLMs can serve as a powerful central component for high-level reasoning and task decomposition, enabling robots to adapt to novel, unstructured environments and understand natural language instructions.
- A modular architecture with specialized LLM agents for different cognitive functions (comprehension, planning, action generation) facilitates complex robot pipeline design.
- Integrating an affordance model driven by an LLM is critical for enabling zero-shot, task-oriented grasping of tools, significantly enhancing a robot's manipulation capabilities beyond simple pick-and-place.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.