lepetitroseau's picture
Update README.md
a443a9d verified
metadata
license: apache-2.0
size_categories:
  - n<1K

🥪 leChef: A Multi-Task Robotic Dataset for the SO 100 Arm with Lateral Mobility leChef is a multi-task robotic manipulation dataset built for the SO 100 robotic arm, enhanced with a custom lateral mobility rail. This additional degree of freedom allows the arm to move horizontally, expanding its workspace and introducing new challenges in coordination and control.

🎯 Dataset Highlights Robotic Platform: SO 100 arm

Mobility Feature: Lateral motion via rail

Number of Tasks: 5

Episodes per Task: 40

Total Episodes: 200

Learning Design: All tasks are designed to be learned by a single ACT model with minimal modifications

🧠 Task Description Each task consists of grabbing a sandwich ingredient and placing it onto a plate, stacking it on top of the previous ones. The five tasks follow the natural construction of a sandwich:

Grab Bread 1 and place it on the plate

Grab Cheese and place it on the plate

Grab Ham and place it on the plate

Grab Salad and place it on the plate

Grab Bread 2 and place it on the plate

This layered structure creates a progressively changing environment across tasks, ideal for studying manipulation under evolving scene configurations.

📌 Purpose The leChef dataset supports research in multi-task learning, manipulation in dynamic environments, and the integration of mobility into fine-grained control. The shared structure across tasks makes it suitable for training a single ACT model with only minor task-specific adjustments.