what is behavior_prompting fr?

Behavior Prompting Policy is research code from a Stanford robotics lab, built to accompany a paper about teaching robots new tasks by showing them a demonstration instead of retraining a model from scratch. The idea is that a robot policy can be prompted with an example, much like showing a person how to do something once, rather than needing thousands of new training examples for every new task. The repository bundles several related pieces of robot learning research together. It includes training for visuomotor policies, meaning models that learn to act based on camera images, with support for following language instructions, following a goal image, or following an in-context demonstration. It also includes support for iPhUMI, a system for training single arm and two arm robots using demonstrations captured on an iPhone, and code for deploying trained policies onto real robots and sensors. The project also extends and evaluates against LIBERO, a common benchmark suite for robot manipulation tasks, adding a version called LIBERO-Gen that can automatically generate new environments, tasks, and demonstrations for training and testing. A separate simulated environment called DrawAnything-Sim is included to test how well a policy can learn 2D drawing tasks. Setup requires cloning with git submodules, opening the project as a VS Code workspace, and creating a mamba (a Conda-like package manager) environment from a provided file, with a second environment needed only if deploying to a physical robot. Detailed setup guides are provided as separate documents for each subsystem, such as iPhUMI, laundry folding, and LIBERO. This is specialized research code intended for people working on robot learning and manipulation, not a general purpose tool. It is released under the MIT license, and the authors ask that users cite their paper if the code is useful in research.