Doctoral Speaking Skills Talk - Benjamin Stoler

— 2:00pm

Location:
In Person - Gates Hillman 8102

Speaker:
BENJAMIN STOLER, Ph.D. Student, Computer Science Department, Carnegie Mellon University
https://benstoler.com/

Autonomous vehicles must navigate among humans in a safe and socially-compliant manner. Current approaches for building and evaluating such systems rely on data-driven techniques; however, a generalization gap emerges, as methods trained in these traditional paradigms are unable to cope with unexpected real-world scenarios. Therefore, we aim to develop improved evaluation settings and methodologies to increase and assess robustness in autonomous driving against these challenges.

For robustness against out-of-distribution, safety-relevant scenarios, we create a hierarchical characterization method which leverages counterfactual probes to find hidden safety-relevant scenarios in large datasets. We then address the induced generalization gap by incorporating the characterizations into downstream trajectory prediction models' inductive biases. Next, for robustness against adversarial, safety-critical scenarios, we develop a reactive, skill-based adversary policy which leverages a learned, multi-faceted criticality objective to perturb existing scenarios. We then train ego policies in a closed-loop manner against these generated scenarios, demonstrating improved downstream ego performance.

This talk concludes by outlining future directions to further advance reliability and safety in autonomous driving. 

Presented in Partial Fulfillment of the CSD Speaking Skills Requirement

Add event to Google
Add event to iCal