College of Science & Engineering
Twin Cities
This research group investigates various aspects of dynamic systems moving in a real or virtual environment (e.g., physical simulations, motion prediction, and crowd flow analysis). Their particular focus is on motion planning, which involves an agent making a prediction of how the environment will likely develop over time, and then finding an optimal plan that respects the predicted development of an environment. For example, if a robot is trying to exit a room, it should try to understand if someone in the room is already moving towards the door and, if so, wait its turn before leaving. This problem is exacerbated by considering motion constraints a robot might have (don't accelerate too much, or turn too fast), and the real-time computational constraints of interacting with people.
Recently, the group's research has taken an increasing focus on data-driven approaches to motion prediction, simulation evaluation, and even planning techniques. Some examples of recent work include:
- Large-scale data analysis human paths when navigating in indoor spaces and buildings
- Developing methods for multi-robot coordination in dense scenarios through reinforcement learning
- Proposing an encoder-decoder neural network framework to provide human like navigation
Research by this group was featured on the MSI website in September 2014: Simulations of Interacting Objects.