Building a quadruped robot for reinforcement learning research

Supervisor: Dr. Rituraj Kaushik (rituraj.kaushik@aalto.fi)

Learning goals:

  • 3D printing robot’s body parts using pre-designed models.
  • Programming the microcontroller to control the dynamixel smart actuator.
  • Communicating with the robot using USB cable.
  • Robot’s pose estimation using Optitrack motion capture system.
  • Robot operating system (ROS).

Project description

The goal of this project is to build the open-sourced quadruped robot called Real-Ant. To start with, the 3D models and the codes for the microcontroller as well as serial port communication are available for the robot. So the first task will be to print the components, assemble the robot and test the available codes on the robot. Then in the second stage, we will modify the robot to make it more robust so that it can tolerate mild shocks and can be run for an extended period of time without the need of tightening the screws or replacing the components. For example, we can think of adding soft legs, cushions etc. to protect the belly and the actuators. In addition, we can also simplify the design so that it can be repaired quickly if required.

Deliverables:

  1. Fully functional reconstruction of the quadruped robot using the open source model and the codes.
  2. Modification of the design of the robot improve its life and reduce the complexity to repair.

References:

  1. RealAnt – https://github.com/OteRobotics/realant
  2. Reinforcement Learning with RealAnt – https://github.com/AaltoVision/realant-rl