Virtual Evaluation of Deep Learning Techniques for Vision-Based Trajectory Tracking
2022-01-0439
03/29/2022
- Event
- Content
- Ground vehicle systems are still largely reliant on traditional control technique deployments. Deep-learning based control-system deployments, are emerging as a viable substitute to more traditional control-system. However, reliability and robustness of deep-learning based controllers in actual deployments and their subsequent verification and validation remains a challenge. This is exacerbated by the need to factor in the uncertainty of the environment as well as the increased number of parameters. Existing literature comparisons of deep-learning vs traditional controllers do not offer structured approaches to performance evaluation and improvement. It is also crucial to: (i) develop a standardized controlled test-environment within which various controllers are evaluated against a common metric; (ii) identify a reference high-fidelity controller (traditional) that can serve as a benchmark. Hence, in this paper, we evaluate deep learning-based controllers by a structured selection of pantheon of evaluation metrics and employ the insights to propose further modifications to the deep learning algorithms. We first evaluate a high-resolution PID controller to serve as a common benchmark and develop the suite of evaluation metrics for a mobile robot to create a point-to-point trajectory within in the simulation environment. Against this backdrop, we then evaluate alternate variants of imitation-learning controllers which use camera image-stream as an input to develop/tune trajectory planning output parameters. This includes: (i) Direct Imitation-Learning approach to inference velocity- and yaw-rate output simultaneously; (ii) Deep Reinforcement Learning; and finally (iii) Parameter Learning approach where deep reinforcement learning is used to tune the gains of an existing traditional (PID) controller. Subsequent deployments of control strategies on actual hardware (in a Sim2Real transition) is anticipated – the developed evaluation metrics can be used as a validation tool when deploying the control strategy on actual hardware.
- Citation
- Salvi, A., Krovi, V., Buzhardt, J., Tallapragda, P. et al., "Virtual Evaluation of Deep Learning Techniques for Vision-Based Trajectory Tracking," SAE Technical Paper 2022-01-0439, 2022, .