Full details and results will be posted following publication.

This project focuses on developing a modular robotics framework that spans hardware prototyping, physics-based simulation, and machine learning. It began with CAD design and experimental testing in a Vicon motion capture environment, where initial control algorithms were validated. The project then transitioned into simulation, adapting real-world control scripts into a scalable MuJoCo environment for dynamic connection and reconfiguration of robotic modules. With the simulation established, the next phase explores machine learning to enable adaptive and autonomous behaviors.

Goals

Prototype and validate modular robotic hardware through CAD design and Vicon-based testing.
Translate real-world control algorithms into a physics-based simulation environment.
Develop robust algorithms for automated module connection, chain formation, and collective control.
Establish a modular software architecture to support extensibility across hardware and simulation.
Advance toward reinforcement learning to enable adaptive and autonomous robot behaviors.

Features

CAD design in Fusion360 and motion capture testing in Vicon.
Control framework and algorithms written in Python.
Physics-based simulation using MuJoCo.
Integration with reinforcement learning workflows (in progress).

No items found.

Specific Contributions

Built a MuJoCo simulation framework that accurately reflects real-life control behaviors.
Implemented user-centered algorithms for module connection and chain formation.
Designed inter-module docking mechanism and internal component housing.
Designed a unified and scalable data architecture for managing module states.
Laid groundwork for reinforcement learning applications to extend adaptability and autonomy.