Aditya Narayan

AGV

Autonomous Ground Vehicle Research Group is a multi-disciplinary research group aimed at building a fully operational self-driving car. I have been a part of the research group since February 2014 in its software team.

My contributions to Eklavya - 3, the last version of Eklavya series that participated in IGVC 2014 include

  • Ability to create varying seeds.

    A seed is a circular path that a robot can take from its current direction. Using seeds, we can restrain out robot from attempting irregular paths with varying curvatures. A seed can be uniquely identified by the position vector of the end point of a seed arc. I developed a script that would list all possible seeds around a point as required, interpret the curvature and accordingly assign velocities to both the wheels to achieve such a curvature.

  • Introduction of a newer *Quick Reflex Planner over the traditional A-seed search planner.

    The Eklavya - 3 employed a LIDAR that fused with the camera module to create a map with 20m visibility around the robot. A seeded A* star, while discovering the entire map, would assume the entire map to be the search space. This meant a performance of 35 FPS, good enough for straight lanes but not adequate enough for tight turns.
    It was observed that the optimum path to a target say 20m away (while constrained to move within a lane) would always involve the optimum path to a target just 5m ahead of the bot, in the direction of heading. Reducing our search space to a quarter of the previous value allowed us to achieve 200 FPS.

The latest iteration of Eklavya, the Eklavya - 4 will compete in IGVC 2015. My contributions include

  • Catkin package for VectorNav VN-200 Rugged.

    The VectorNav VN-200 Rugged is a all-purpose INS solution employed in Eklavya series. I developed a package to publish the IMU and INS data as Sensor/Imu.msg. The package also employed quick post-processing to eliminate gravity and random walk errors.