How glnea/rpent works
RPent is an extensible framework for embodied agents, spanning simulated manipulation benchmarks and Franka hardware. The main workflow starts with an operator task, then uses a configured planner and robot-specific tools to observe and act through an environment service; memory and human interaction support the recursive agent loop. The README also documents evaluation and memory-distillation capabilities. The graph combines those documented product branches with sampled CLI, RPC, and environment-client behavior; several unsampled subsystem links are therefore left out.
Main entry points: main.py RPent CLI · common.py Robot Tools · toolkit.py Toolkits · manager.py Memory Manager
Components 20
Agent Control
- RPent CLI
rpent/cli/main.py - Robot Specs
rpent/robots/robot_spec.py - Planners
rpent/planner - Toolkits
rpent/tools/toolkit.py - Robot Tools
rpent/tools/common.py
Robot Runtime
- Perception Services
rpent/robots/components - Environment Facades
rpent/robots/components/env_facade_base.py - Environment Clients
rpent/robots/components/env_client_base.py - Benchmark Adapters
robots/libero/env_client.py - Franka Adapters
robots/franka/env_client.py - VLA Client
rpent/robots/components/pi05_vla_client.py - Simulator or Robot
Memory Learning
- Memory Manager
rpent/memory/manager.py - Memory Corpus
- Memory Tools
rpent/memory/tools.py
Operator Interaction
- Dashboard
rpent/dashboard/server.py - Human Input
rpent/tools/human_in_the_loop.py
Evaluation
- Evaluation Results
rpent/evaluation/result.py - Flywheel Export
rpent/flywheel/export.py
Other components
- Operator
Connections 22
- Operator → RPent CLI: sets task
- RPent CLI → Robot Specs: loads configuration
- RPent CLI → Planners: runs solve
- RPent CLI → Toolkits: supplies tools
- Planners → Toolkits: invokes tools
- Toolkits → Robot Tools: dispatches calls
- Robot Tools → Environment Facades: requests actions
- Environment Facades → Environment Clients: uses client
- Environment Clients → Benchmark Adapters: specializes
- Environment Clients → Franka Adapters: specializes
- Benchmark Adapters → Simulator or Robot: RPC calls
- Franka Adapters → Simulator or Robot: RPC calls
- Robot Tools → Perception Services: requests perception
- Robot Tools → VLA Client: requests actions
- RPent CLI → Memory Manager: configures memory
- Memory Manager → Memory Corpus: reads and merges
- Memory Tools → Memory Manager: accesses corpus
- RPent CLI → Dashboard: starts dashboard
- Dashboard → Operator: receives interaction
- Toolkits → Human Input: offers approval
- RPent CLI → Evaluation Results: finalizes run
- Memory Manager → Flywheel Export: supports distillation