Interaction-driven Behavior Prediction and Planning for Autonomous Vehicles

Photo credit: Unsplash

This workshop will be held as full day workshop at the 28th IEEE International Conference on Intelligent Transportation Systems (ITSC), at Gold Coast, Australia

Topics

The topics of interest of the workshop include, but are not limited to:

Cooperative and comprehensible motion planning
Probabilistic decision making and motion planning (including MDPs, POMDPs, MMDPs)
Probabilistic behavior prediction (with help of semantic high-definition maps)
Second-order effects in heavy interactive scenarios
Evaluation and benchmarking of the aforementioned topics

Workshop Content

Research on Automated Vehicles has experienced vast progress over the last decades. Today, first prototypes are sufficiently safe to drive on selected roads in public traffic. Nevertheless, safety comes at the price of overly conservative behavior, leading to inconvenient situations, for example, at unprotected left turns or merging scenarios. Presumably, the main reasons for this behavior include:

Errors in the prediction of other traffic participants, especially in interactive scenarios
Lack of probabilistic considerations in motion planning

Comfortable Automated Driving

While safety should never be put at risk, worst-case behavior of others should not be the default for the motion plan of an automated vehicle. Rather, with a safe reaction to such worst-case behavior always in reserve, the intended trajectory should be comfortable, less conservative, and thereby potentially closer to human expectations. Proposal and exchange of these kinds of approaches is the first aim of the workshop.

Multimodal Behavior Prediction

For such behavior, sophisticated behavior prediction approaches for other traffic participants are necessary, going beyond constant velocity assumptions. Predictions must be probabilistic and allow for maneuver options for other vehicles. Often, there is not “the right prediction,” but many. The choice is influenced by destinations, individual driving behaviors, and potentially even the driver’s mood. Thus, a simple evaluation against a ground truth is not possible. Prediction approaches (including machine learning), and proposals for their evaluation, are the second main goal of this workshop.

Comprehensible Automated Driving

For motion planning in highly interactive scenarios, a “ground truth” or “best option” may not exist. To be comprehensible and predictable for other road users, a good plan should be a subset of an expected prediction for a vehicle in the same situation. The combination of planning and prediction, including evaluation and benchmarking, is the third aim of the proposed workshop.

Effects of Automation on Traffic

Data-driven predictions can end up being implicitly conditioned on second-order effects. For example, seeing a recording vehicle or no driver in an autonomous car can influence traffic participant’s decisions. Fixed settings in automated functions, such as safe distances, can influence the traffic flow on highways. While this can potentially introduce a distribution shift for prediction algorithms, it could also be leveraged to purposefully shape traffic. We invite approaches investigating these second-order effects, propagating in highly interactive scenarios.

Tentative Schedule (Paper talks will be 20min + time for questions and speaker change. Exact times may vary by +- 5 min)

Time	Name and Affiliation	Type of Presentation	Presentation Title
08:30	Welcome Ceremony
08:30 - 08:55	Enli Lin, Tsinghua University, China		Learning Complex Urban Traffic Dynamics based on Interaction Decoupling Strategy
08:55 - 09:20	Fabian Konstantinidis, Karlsruhe Institute of Technology, Germany		Conditional Prediction by Simulation for Automated Driving
09:20 - 10:00	Prof. Ehsan Hashemi, University of Alberta, Canada	Keynote	Game-Theoretic based Motion Planning for Safe Human Autonomy Interaction
10:00 - 10:30	Morning Break
10:30 - 11:00	Anjian Li, Princeton University, USA.		Predictive Planner for Autonomous Driving with Consistency Models
11:00 - 11:30	Marc Kaufeld, TU Munich, Germany		MP-RBFN: Learning-based Vehicle Motion Primitives using Radial Basis Function Networks
11:30 - 12:00	Daniel Grimm, FZI, Germany		Goal-based Trajectory Prediction for improved Cross-Dataset Generalization
12:00 - 12:30	Divya Thuremella, Oxford Robotic Institute, UK		Long-Tailed Learning for Trajectory Prediction
12:30 - 13:30	Lunch Break
13:30 - 14:00	Yuan Gao, TU Munich, Germany		From Words to Collisions: LLM-Guided Evaluation and Adversarial Generation of Safety-Critical Driving Scenarios
14:00 - 14:30	Nihed Naidja, CentraleSupelec, France		Generalized Nash Equilibrium-Based Decision Making for Autonomous Vehicles
14:30 - 15:00	Francesco Prignoli, University of Bologna, Italy		From Regulations to Strategy: Game-Theoretic Motion Planning for Autonomous Racing
15:00 - 15:30	Eric Roy-Almonacid, Universitat Politècnica de Catalunya, Spain		Games of ordered preference for mobility systems
15:30 - 16:00	Afternoon Refreshments
16:00 - 16:30	Panel Session
16:30 - 16:40	Closing Ceremony
16:40 - 17:30	Poster Session