Repolode
Submit
RLinf

RLinf/RPent

An open framework for building embodied agents that use LLM planners, VLA manipulation models and memory to run robot tasks in simulators and on real robots.

View on GitHubWebsite
Stars
1,554
Forks
136
Open issues
37
Last commit
6 hours ago

What is it?

RPent (Recursive Physical Agent) is an agentic infrastructure for physical robots. Instead of tying the system to one foundation model, it combines perception, reasoning, memory, execution and self-evolution components into a single physical agent, built on service-oriented, standardized and composable design principles.

The framework lets a planner such as Claude Code or Codex decide how to break a task into steps, while VLA models like Pi0.5, RLDX-1 and LingBot-VLA execute the low-level manipulation primitives. It supports LIBERO-PRO, RoboCasa, RoboTwin and RoboDojo simulators, and Franka and dual-Franka real-robot setups are listed as supported, with SO-101 and YAM marked as planned.

RPent ships a CLI for running predefined tasks, an interactive mode for steering the agent from a terminal, and a live dashboard that streams agent reasoning, camera views and the action timeline. Documentation is available in English and Chinese.

Key features

  • 01Pluggable planners, including Claude Code, Codex and custom planners.
  • 02Supports VLA models such as Pi0.5, RLDX-1 and LingBot-VLA as action primitives.
  • 03Runs on LIBERO-PRO, RoboCasa, RoboTwin and RoboDojo simulators.
  • 04Supports Franka and dual-Franka real-robot setups.
  • 05Interactive CLI mode for steering the agent during a run.
  • 06Live dashboard with agent reasoning, camera views and action timeline.

Use cases

  • Benchmarking long-horizon manipulation methods on LIBERO-PRO or RoboCasa.
  • Researching self-evolving or memory-guided embodied agents.
  • Prototyping agentic robot control on Franka arms before production deployment.

Topics

#embodied-ai#robotics#agentic-ai#vla#robot-simulation#reinforcement-learning#llm-agents#embodied

FAQ

What is RPent?
An open framework for building embodied agents that use LLM planners, VLA manipulation models and memory to run robot tasks in simulators and on real robots. RPent (Recursive Physical Agent) is an agentic infrastructure for physical robots. Instead of tying the system to one foundation model, it combines perception, reasoning, memory, execution and self-evolution components into a single physical agent, built on service-oriented, standardized and composable design principles.
Is RPent free to use? What license does it have?
RPent is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is RPent still maintained?
Actively maintained. The last commit to RPent was on Oct 19, 2026.
What can I use RPent for?
Benchmarking long-horizon manipulation methods on LIBERO-PRO or RoboCasa. Researching self-evolving or memory-guided embodied agents. Prototyping agentic robot control on Franka arms before production deployment.
Where can I get RPent?
The source code and installation instructions are on GitHub: https://github.com/RLinf/RPent

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects