Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware
Paper • 2304.13705 • Published • 7
NexArm ACT: Pick and Place
This model is trained with ACT (Action Chunking with Transformers) on a Hiwonder NexArm 6-DOF robotic arm for a pick-and-place task.
Training Details
| Item | Value |
|---|---|
| Policy | ACT (Action Chunking with Transformers) |
| Dataset | 31 episodes, ~27K frames |
| Cameras | 2× USB cameras (front + wrist), 640×480 |
| Training Time | ~29K steps on NVIDIA RTX 5070 Ti |
| Final Loss | ~0.057 |
Usage
from lerobot.policies import ACTPolicy
from lerobot.rollout import rollout
policy = ACTPolicy.from_pretrained("CypherChen/NexarmControlModel")
# Run inference on your NexArm robot
rollout(policy, robot_type="nexarm_follower")
Hardware
- Robot: Hiwonder NexArm (ESP32 + AT32 dual-chip)
- Servos: HX-series serial bus servos
- Leader arm: Drag-to-teach teleoperation
- Follower arm: Executes learned policy
License
Apache 2.0
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support