In order for autonomous systems to be integrated into our daily life, we cannot let robots perform in isolation. Robots need to interact with various intelligent agents including humans and other autonomous systems in complex dynamic environments. There are many instances of such interactions. To name a few, an autonomous car may need to interact with human drivers and other autonomous cars on the road. A team of robots may need to collaborate to manipulate a heavy object in space. Teams of robots may need to compete in a pursuit-evasion game. We are interested in developing control and planning algorithms that enable robots to achieve safe and efficient interactions with other agents.
Interactive Motion Planning
One of the key aspects for achieving safe and efficient robotic interactions is to enable robots to reason about how their presence will affect other agents, i.e. robots need to reason about their interactions with other intelligent agents as well as their impact on the overall system. For instance, consider a team of service robots carrying an object while avoiding a human. Every robot needs to reason about how its actions will affect its teammates and the overall objective of the team as well as how its actions will affect the human. We are interested in developing motion planning algorithms that capture such dynamics and enable robots to navigate in a space that is shared by other agents.
Roles in Robotic Interactions
Human beings have the amazing ability to organize themselves into teams to accomplish difficult tasks. This organizational behavior often takes the form of dividing responsibilities among different individuals as different roles in the group. This can be seen in a large variety of activities ranging from sports (eg. soccer or basketball) to mundane tasks (eg. group projects, collaboratively moving objects). With the prospect of deploying robotic systems in environments shared with humans, it has become more and more important for us to equip robotic systems with the ability to actively engage in interactions with other group members. One of the properties that a robot can exploit is the role composition of the group. We are interested in building a tool that can predict the role composition in multi-agent teams given a description of the environment and the agents’ preferences.