Graduate Research Assistant
Ontario, Canada
• Developed Vision Augmented Vehicle State Estimation with Fault Tolerance prototype for localization of the Autonomoose vehicle. Objective of the prototype was to create accurate and reliable data estimates to be used for different autonomous processes, like localization. The prototype uses the Extended Kalman Filter as the primary estimation approach while being inspired by approaches like FKF to increase estimation robustness. The prototype also uses pose information from Visual Odometry to reinforce the State vector formed by pose information from conventional sources GPS, INS, and wheel odometer. This reinforcement boosts estimation efficiency of the vehicle estimates, especially in conditions of wheel slip, INS errors, and GPS void regions. Fault tolerant strategies intelligently added to mitigate the effect of faults in the data stream, thus achieving higher data robustness and lowering estimation error. A 7 degree of freedom non-linear vehicle model is integrated in to the EKF improving its prediction capability. • Achieved higher data robustness and estimation efficiency by augmenting conventional sensors with Visual Odometry (VO) • Improved state estimates and reduced estimation error by implementing fault tolerance methodology to the estimation architecture• Explored multiple degrees of freedom for a vehicle model• Developed vehicle model prototype on MapleSim to integrate it with VREP simulation platform for replicated locomotion of the simulated vehicle• Developed health monitoring node to track the health of the data messages throughout the estimation and fault tolerance chain• Key member of autonomous architecture design development: fail safe operations, necessary redundancy, health monitoring• Key words: C++, ROS, Node, Nodelet, EKF, Harris, localization, Sobel, FAST, RANSAC, 8 Point Algorithm, Jacobian, Vehicle Model, GPS, INS, IMU, sensor fusion, non linear systems, camera systems
