ACT Real PickOrange is a model available on Hugging Face that provides an ACT policy for real-world robotic pick-and-place tasks, specifically for the SO-101 pick-orange scenario. The model is developed using a process where training begins in simulation, using the Isaac Lab environment (LeIsaac-SO101-PickOrange-v0), to achieve the best performance on a simulation leaderboard. This checkpoint is then used as a starting point (sim-warm-start) and further fine-tuned on real hardware through 30 teleoperation episodes, amounting to 5000 additional training steps. The result is a policy intended for manipulation tasks with real robotic arms, focusing on the challenge of picking oranges in a physical setting. The model is distributed under the Apache 2.0 license. Instructions for using the model with libraries such as LeRobot, as well as with platforms like Google Colab and Kaggle, are provided. No further details about intended users, broader capabilities, or additional features are specified in the available evidence.
In the Other AI space, ACT Real PickOrange takes a focused approach. It focuses on providing a trained AI policy for real-world robotic pick-and-place operations. ACT Real PickOrange is an open-source project aimed at robotics researchers and engineers. The project is open source (Open Source). It runs on the web and the command line.
Behind ACT Real PickOrange is wsagi, and the product first shipped in 2025. The project is developed in the open on GitHub with 47 commits in the last 90 days. Across PulseGate's embedding index, ACT Real PickOrange has few near neighbours, marking it as relatively distinct. Among its 4 catalogued features are robotics policy, pick-and-place, and open source.
Latest indexed changes and source events
wsagi/ACT-Real-PickOrange discovered by the PulseGate indexer
Other apps tracked under the same category.
