Autonomous Driving in Mixed-Traffic Urban Environments

Abstract:

This study attempts to address a range of problems associated with autonomous navigation in real-life urban traffic, most recently including controller design and safety, secure communication, mapping, pedestrian behavior, testing and intersection access. On the control side, researchers investigated safety verification for automated controllers using reachability analysis methods, on cruise control and collision avoidance controllers; and designed intersection-access controllers through adaptive real-time traffic lights utilizing vehicle-to-infrastructure (V2I) technologies. For collaborative mapping, both Kalman filters and scan matching methods were utilized and multi-tier traffic simulations were conducted to evaluate the mapping performance. In order to address the human side of the mixed-mode traffic, pedestrian motion models were studied and a mixed model using portions of social--force methods and agent--based methods was developed. This model was shown to display emergent and realistic behavior that was not inherently built into the model, through mixtures of various dynamics. The researchers continued to investigate sensory data processing and reliable/secure sensor fusion methods in order to address data-security concerns, and a trust-aware particle filter was designed, developed and simulated to achieve a higher-confidence situational awareness for collaborating vehicles in traffic. Semi-virtual and scaled-down test setups/scenarios were developed to leverage the lower cost, safe and repeatable nature of these tests in addition to full-scale, outdoor vehicle experiments. These supplemental test and evaluation methods were shown to be beneficial during the development and for final full-scaled demonstrations.