Welcome to RPent#
RPent (Recursive Physical Agent) is an open-source framework for building embodied agents that continually evolve through recursive interaction with the physical world. It supports different foundation models and integrates perception, reasoning, memory, execution, and self-evolution in a unified agent framework. Through ongoing interaction, agents reflect on and adjust their behavior, accumulate experience, and develop new capabilities beyond their initial design.
Install RPent, run a LIBERO-PRO task, and inspect the result.
Learn how planning, perception, actions, and memory work together.
Use published experience or build memory through exploration.
Compare reported success rates, runtime, and token costs.
Watch dual-arm Franka tasks and follow the deployment guide.
Understand the execution flow and add robots, tools, or planners.
Choose an environment: LIBERO, RoboCasa365, or RoboTwin. For robot deployment, see Single-Arm Franka and Dual-Arm Franka; YAM has a task demo.