Drone Autonomy Using Deep Reinforcement Learning
Avideh Zakhor, Professor
Electrical Engineering and Computer Science
Applications for Spring 2024 are closed for this project.
In this project, we will develop methods to enable a drone to fly effortlessly among obstacles in outdoor environments such as forests. Our testing site is Cesar Chavez park in Berkeley Marina as well as Richmond Field Station. You will be working with an actual drone e.g. Bit Craze, or DJI Mavic Air, or DJI Matrice 300. Will be developing Deep Reinforcement Learning Algorithms to teach the drone to get to destination while avoiding obstacles.
Role: * Develop RL algorithms
* Verify RL algorithms in Simulation e.g. Flightmare or IssacGym
* Test the RL algorithms on actual drone.
* Opportunity to develop algorithms e.g. for super lightweight drone e.g. Bit Craze with a simple AI Deck, or for a more capable drone Matrice 300 with Nvidia Jetson, or for a mid size drone e.g. Mavic Air 2.
* Opportunity to work on indoor GPS denied regions, or outdoors with GPS.
Qualifications: * Must have taken or currently be taking CS285
* Must have taken basic deep learning classes e.g. CS189 or CS 182.
* Prior experience with Pytorch and deep learning algorithms
* Familiarity with ROS is a plus.
* Familiarity with visual odometry and SLAM is a plus.
Hours: 12 or more hours
Related website: http://www-video.eecs.berkeley.edu/~avz
Engineering, Design & Technologies