Deployment Study Imitation Learning · 2026
ALLEX ACT
ACT Training and Real-Robot Deployment on ALLEX
An end-to-end imitation-learning study for a 22-DoF ALLEX arm-hand system, from 50 Hz demonstrations to ACT training and real-robot task execution.
01
What I Did
- Collected and curated approximately 85 teleoperated demonstrations at 50 Hz for a 7-DoF arm and 15-DoF hand.
- Converted raw demonstrations into a LeRobot dataset and trained an ACT policy for 1,000,000 steps on an RTX 5080.
- Validated the learned policy in Isaac Sim through ROS 2 before moving to the physical robot.
- Demonstrated a complete real-robot sequence that collected three objects and placed them in a box.
The current result is a proof of concept; repeatability and success-rate evaluation remain future work.
02
Topics · Project Details
Focus
Demonstration collection, ACT training, simulation validation, and real deployment
Topics
ALLEXACTImitation LearningLeRobotIsaac Sim
Stack
PythonROS 2LeRobotACTIsaac SimPyTorch
Details
- Track
- Imitation Learning
- Stage
- Graduate
- Year
- 2026
- Status
- Current