Unprecedented progress in computing and machine learning techniques have made autonomous vehicles (AVs) practical, sharing the road with humans. The behaviors of humans, however, are naturally full of uncertainties with influences from not only individual preferences but also social impacts. Moreover, the upstream modules such as perception and prediction also produce uncertainties with their results. This makes the behavior generation and decision-making problem of AVs extremely challenging to achieve safe and efficient interactions with humans.

There have been lots of approaches proposed to address the above challenges. Strategies differ from ways to generate behavior (e.g., imitating experts’ behaviors or computing behaviors via optimization), to reward/cost function design, to interaction formulation, and to the formats about how uncertainties from upstream results are incorporated. Different approaches have their unique advantages in different domains, and all these variances lead to autonomous vehicles with different safety and efficiency levels. Thus, a discussion about potential performance of above mentioned state-of-the-art approaches for a potential uniform framework covering different driving scenarios with safer and more efficient AVs are strongly desired.

Researchers in related areas from both, academia or industry are invited to submit extended abstracts (at least 3-pages long) or full papers to be presented in spotlight presentations and a poster session. The topics of interest include but are not limited to:

• Decision-making under uncertainty, such as POMDP;

• Reinforcement learning (RL) and inverse reinforcement learning (IRL);

• Imitation learning;

• Deep generative models such as variational auto-encoder (VAE), generative adversarial
  network (GAN) and their applications;

• Relational reasoning in autonomous driving;

• Model uncertainty in deep learning;

• Formulation of interactions and mutual/group influences;

• Decision-making and behavior generation with uncertainties;

• Game-theoretic formulation;

• Evaluation of different approaches on a variety of scenarios;

• Uniform platform across multiple scenarios;

• Benchmark metrics to better quantify the performance of behavior generation and
  decision making;

deadline: March 14th, 2020