Building a quadruped robot for reinforcement learning research

Basic Information

Students: Antti Sippola, Eric Hannus, Jed Muff, Jere Vespä, Julius Mikala, 

Project manager: Jed Muff

Instructor: Rituraj Kaushik

Other advisors: - 

Starting date: 24/01/2022 

Completion date: 03/06/2022 

The project aimed to build and improve a quadruped robot that will be used for reinforcement learning research. The quadruped robot built is based on the open-source Real-Ant designed by Ote Robotics at Aalto University. With the previous design, issues with robustness and usability were identified. These are related to the screw fittings and the repairability of the design. Design improvements included soft attachments to dampen damage to the robot, proper board mounting for easy construction, access hatches for ease of use and repairability and a stand for a motion tracking marker (used in reinforcement learning).
Along with an improved robot design, some code was developed that utilized DYNAMIXEL AX12 actuators to produce walking gaits and gestures the quadruped robot could use for initial functionality testing. This project involved documenting the final designs and the code developed.

Overall the project was a success, and all objectives were achieved.