Understanding Others: Robot Learning in Interactions

We are motivated by the problem of building autonomous robots that are able to work collaboratively with other agents, such as human co-workers. One key attribute of such an autonomous system is the ability to make predictions about the actions and intentions of other agents in a dynamic environment - both to interpret the activity context as it is being played out and to adapt actions in response to that contextual information.

I will present recent results addressing questions of how to efficiently represent the hierarchical nature of activities, how to rapidly make inferences about latent factors, such as hidden goals and intent, and how to make optimal decisions in interactions without explicit prior coordination with unknown partners.

Firstly, I will describe a procedure for topological trajectory classification, using the concept of persistent homology, which enables unsupervised extraction of certain kinds of relational concepts in motion data. One use of this representation is in devising a multi-scale version of Bayesian recursive estimation, which is a step towards reliably grounding human instructions in the realized activity.

I will then describe work with a human-robot interface based on the joint use of mobile 3D eye tracking and vision for intention inference. We achieve this through the use of a probabilistic generative model of fixations conditioned on the task that the person is executing. Using preliminary experimental results, I will discuss how this approach is useful in the grounding of plan symbols to their detailed appearance in the environment.

Finally, I will present a model and algorithm for optimal decision making in interactions with unknown partners, based on the use of “policy types” within an incomplete information game, combining the benefits of Harsanyi’s notion of types and Bellman’s notion of optimality in sequential decisions. Using results from human-machine experiments, I will show how this algorithm achieves a better rate of coordination than alternate multi-agent learning algorithms.

Speaker: Subramanian Ramamoorthy, Univ. of Edinburgh