See & Spray Technology

Ottava Surgical Robot

Indoor autonomous mobile robots

The problem of navigation in a partially observable environment is common to all types of robots. In order to work out a solution, one needs to do online state estimation, plan and control the robot. This involves a mix of localization, motion and path planning and a model predictive control framework. I worked on all of such concepts during my Masters for a point robot and a legged robot. Since online belief space planning (another term for simultaneous localization and planning) is still an unsolved problem in legged robots due to large dimensionality, I tried to reduce that by using information only from IMU and forward kinematics to do state estimation.

My role was to perform Visual SLAM (SFM) using stereoscopic cameras attached to the drone. I used the ORB SLAM technique to do the same. I was also instrumental in tasks like object recognition and detection.