Bebop Navigation
collision cone based optimization for dynamic obstacles
Robotics Research Centre, IIIT Hyderabad
This project comprised of implementing dynamic obstacle avoidance on Bebop drone. The goal reaching was an optimization scheme with obstacle avoidance done using collision cone constraints. This was done for holonomic and non-holonomic setting in two different ways.
Holonomic Setting
For this setting a new variant of velocity obstacle was implemented - Inverse Velocity Obstacle where we construct the collision cone in the ego-centric frame which showed improvement for real time implementation. We eliminate the need to know the robot position and velocity for obstacle avoidance - thus eliminating the associated noise. It was shown to work for multi-agent setting as well. This work was published at AIR 2019.
Non-Holonomic Setting
In this the drone was constricted to move like a fixed-wing aircraft. To implement this, the optimised path was decomposed into dubin paths. This scheme was tested in real life. All the detailed video results are available here
