Event
Ph.D. Research Proposal Exam: Muhammad Faizan Tariq
Tuesday, February 1, 2022
1:00 p.m.
Zoom - Link: https://umd.zoom.us/j/95379940331?pwd=cG10VGZKalBLL25WdWlrc2RFTm1EZz09
Maria Hoo
301 405 3681
mch@umd.edu
ANNOUNCEMENT: Ph.D. Research Proposal Exam
Name: Muhammad Faizan Tariq
Committee:
Professor John Baras (Chair)
Professor Eyad Abed
Professor Dinesh Manocha
Date/time: Tuesday, February 1st, 2022 - 1:00 p.m.
Location: Zoom - Link: https://umd.zoom.us/j/95379940331?pwd=cG10VGZKalBLL25WdWlrc2RFTm1EZz09
Title: Safe Navigation of Autonomous Vehicles in Structured Mixed-Traffic Environments
Abstract:
The main motivation driving the autonomous vehicle research forward is the promise of improved road safety resulting from the elimination of accidents caused by human error. To that end, it seems rather unlikely that all the human-driven vehicles will be replaced by autonomous vehicles in the near future. The more plausible scenario is that the autonomous vehicles will gradually be introduced on public roads and highways in the presence of human-driven vehicles, leading to mixed-traffic scenarios. Therefore, it is only practical to focus on research avenues addressing mixed-traffic settings. This, however, presents numerous challenges, especially in regards to robustness to the variations in human driving patterns. Furthermore, the overall safety of the autonomous vehicle is highly contingent upon various uncertain factors such as environmental conditions, measurement accuracy, perception, prediction etc. Therefore, an accurate risk assessment in various encountered scenarios is necessary in order to guarantee safe operation of the vehicle while yielding performance improvements (in space and time) making the algorithmic design task even more challenging.
In this thesis, we aim to tackle the navigation (planning and control) problem with a multi time-scale predictive control framework. We first identify and exploit the different time scales involved in the navigation architecture (e.g. long vs short prediction, planning and control horizons). Then, we pursue systematic complexity reduction in the various algorithmic modules (data and computation) in order to gain immediate performance improvements in regards to inference and action/response delay minimization. This further leads to real-time situation assessment, computation and action/control, which are some of the key requirements for algorithms deployed on autonomous vehicles. We further utilize formal methods involving finite time temporal logic in conjunction with reachability analysis in order to provide safety guarantees, and also explore risk sensitive planning methodologies to ensure robustness in autonomous vehicle operation. Notably, the algorithms proposed in this thesis ensure that the safety of the overall system is a fundamental constraint built into the proposed methods. Distinctive features of the proposed approaches include real-time operation capability, reliable trajectory prediction of on-road entities, safety guarantees, and some degree of robustness to the behavior variability of on-road agents.
In regards to the verification and validation procedures for automated driving tasks, special consideration needs to be placed on utilizing realistic driving environments that consider mixed-traffic, unpredictability of human driven vehicles and limitations in computation resources. In the spirit of emulating real-world situations, three closely related scenarios, implemented in realistic simulation environments, are chosen as bases for validation of the proposed architectures. We employ overtaking on a bidirectional road scenario to validate the receding horizon optimization based approach, maneuvering in a dense highway environment to validate the risk-sensitive planning approach and merging on a highway junction scenario to validate the formal methods based approach. Although seemingly distinct, the fundamental linkage between these scenarios is the overarching need for real-time operation, reliable motion prediction of on-road entities, safety guarantees and robustness to uncertain factors, which is a prevalent theme in our research.
