Humanoid Robotics

Trajectory Optimization using Non-Linear Optimization

Description:
This project explores nonlinear trajectory optimization for the Atlas Humanoid using Centroidal Dynamics and Full Joint Kinematics. We strived to understand the planning and control pipeline which allows Boston Dynamics to execute acrobatic maneuvers. At the conclusion of the project, we showed acrobatic trajectories generated using our pipeline and developed visualizations and intuition to introspect non-linear optimization problems.

Software:
Trajectory Generation in Matlab (see TrajectoryGeneration.m)
Visualization using PyPnC, PyBullet, ROS2 (see atlas_ros)

Project Materials:
Final Presentation
Final Presentation Slides


Machine Learning for Bipedal Locomotion

Description:
In this project, we desired to better acquaint ourselves with the leading edge of bipedal locomotion. Two approaches of interest were explored, including Pure Reinforcement Learning for bipedal walking, and applying learned inertia models to Model Predictive Control (TOWR+).

For two novices to learning techniques, this project was more concerned with exploring new areas and improving our knowledge that it was producing state of the art results.

Project Materials:
Final Presentation
Final Presentation Slides

Bipedal Walking using Divergent Component of Motion

Description:
This assignment leverages the PyPnC environment created by the Human Centered Robotics Lab out of UT Austin. Here, we explore a planning algorithm for legged locomotion known as Divergent Component of Motion. Using tools in PyPnC, walking is demonstrated on Atlas in a PyBullet simulation.

Software:
DCM Planner

Project Materials:
Project Report