Learning a Digging Policy with a Differentiable Granular Material Simulator

Project description

This project explores the use of an existing differentiable granular material simulator for robotic digging. The main objective is to reproduce and understand the released simulation framework and adapt it to a simplified digging task. A basic reinforcement learning method will be used to train a bucket or shovel to interact with and remove granular material. The focus of the project is primarily on setting up and evaluating the simulator for robotic digging, using a basic reinforcement learning method for policy training.

The project will investigate the definition of the digging environment, observations, actions, and reward functions, and evaluate whether meaningful digging behaviours can be learned in simulation. If time permits, simple variations in soil properties, initial terrain, or digging objectives can also be explored.

Reference: https://github.com/IanYangChina/DDBot-IEEE-TRO-2025

Annonsuppgifter

Annonsör: Örebro universitet

Ansök senast: Löpande

Annonskategori: Examensarbete, praktik, uppsats

Intresseområde: Data och IT

Kontaktperson: Johannes Stork (Associate Professor) johanners.stork@oru.se