Franka Lift Object Task

This example demonstrates the complete workflow for reinforcement learning-based object lifting using the Franka Panda robot in Isaac Lab - Arena, covering environment setup, policy training with RSL-RL, and closed-loop evaluation.

Task Overview

Task ID: lift_object

Task Description: The Franka Panda robot learns to grasp and lift objects to target positions through reinforcement learning. The task uses a command-based goal specification, where the RL agent learns to reach sampled target poses.

Key Specifications:

Property	Value
Tags	Table-top manipulation
Skills	Reach, Grasp, Lift
Embodiment	Franka Panda (9 DOF arm + 2 DOF gripper)
Scene	Table with ground plane and lighting
Objects	Configurable (default: dex_cube)
Policy	RSL-RL PPO (learned from scratch)
Training Method	Reinforcement Learning (on-policy PPO)
Physics	PhysX (50Hz @ 2 decimation)
Closed-loop	Yes (50Hz control)
Command Space	Target position [x, y, z] relative to object initial pose
Training Time	~6 hours (12,000 iterations on 512 environments in a A6000)

Workflow

This tutorial covers the pipeline for creating an RL environment, training a policy using RSL-RL, and evaluating the trained policy in closed-loop. A user can follow the whole pipeline, or can start at any intermediate step by using the provided checkpoints.

Prerequisites

Start the isaaclab docker container

./docker/run_docker.sh

We store models on Hugging Face, so you’ll need log in to Hugging Face if you haven’t already.

hf auth login

You’ll need to create folders for logs, checkpoints, and models:

export LOG_DIR=logs/rsl_rl
mkdir -p $LOG_DIR
export MODELS_DIR=models/isaaclab_arena/reinforcement_learning
mkdir -p $MODELS_DIR

Workflow Steps

Follow the following steps to complete the workflow:

• Environment Setup and Validation

• Policy Training

• Closed-Loop Policy Inference and Evaluation