We present CoverNet, a new method for multimodal, probabilistic trajectory prediction for urban driving. Previous work has employed a variety of methods, including multimodal regression, occupancy maps, and 1-step stochastic policies. We instead frame the trajectory prediction problem as classification over a diverse set of trajectories. The size of this set remains manageable due to the limited number of distinct actions that can be taken over a reasonable prediction horizon. We structure the trajectory set to a) ensure a desired level of coverage of the state space, and b) eliminate physically impossible trajectories. By dynamically generating trajectory sets based on the agent's current state, we can further improve our method's efficiency. We demonstrate our approach on public, real world self-driving datasets, and show that it outperforms state-of-the-art methods.

Human-Robot Collaboration (HRC) aims to develop robots that provide assistance to human workers while performing physical tasks. Such assistance comes in the form of supportive behaviors that are different from the actions part of the task, and that are meant to help a human worker more effectively accomplish the task. Learning how to provide useful behaviors that are tailored to a human peer represents a difficult challenge. This is due to the need of large amounts of training data in the form of real world observations that include information about such preferences. This data needs to encode not only the structure and progression of the task, but also the different workers' preferences with respect to when and what assistance the robot should provide. Our work separates the challenge of learning a model of the task (which requires a large amount of training data) from that of learning supportive behavior preferences for the interaction (which has obvious restrictions for the number of user-provided demonstrations to which we have access). We first learn a hidden Markov model (HMM) from a training set consisting of observed human workers performing the considered task in simulation. We then use this model to predict, while observing the human peer, what supportive behaviors a robot should offer throughout the task. Building upon the hidden state representation, our system is able to learn the supportive behaviors based on as few as five user-annotated demonstrations, learning a personalized supportive behavior model. We evaluate our system on a user study with 14 participants, and show results on par with human-level prediction for the task.

Current state-of-the-art robotic systems deployed in industry work in isolation from humans and do not allow for collaboration. Developing a robot that can work side-by-side with a human presents the advantage of allowing both the robot and the human worker to focus on the task each is best suited for, while assisting one another as needed. For the robot to provide assistive behavior to a human co-worker, it needs to learn what actions it should perform at each time step depending upon the state of the task. Such assistive actions are not intended to simply contribute to the completion of a particular task by instructing the robot to work on subtasks in isolation from the human worker; rather they are meant to help the worker complete the task more efficiently. As such, employing standard policy search or task and motion planning techniques is not sufficient to discover the supportive types of actions my system seeks to offer based on accurate estimations of the current task state. To this end, my research focuses on investigating policy search within hierarchical tasks that allow for two main abilities, namely helping the human co-worker more effectively complete a task and taking communicative actions that reduce state estimation uncertainty by asking the worker direct questions. The policy dictates what action the robot should take at each time step, based on inputs from a motion capture system providing observations about the configuration of the person's hands relative to the objects needed for accomplishing the task, as well as the person's answers to any questions posed by the robot.

Mimicry, the automatic imitation of gestures, postures, mannerisms, and other motor movements, has been shown to be a critical component of human interaction but needs further exploration in human-robot interaction. Understanding mimicry is important for building better robots, learning about human categorization of robots in social ingroups/outgroups, and understanding social contagion in human-robot interaction. We investigate the extent to which humans will mimic a robot during the task of describing paintings by comparing the time participants put their hands on their hips before and after observing a robot cue that behavior. We observed no significant difference in participants' hands on hips time before and after the robot's cue. However, we did find that some participants performed the behavior more after the robot's cue while others performed it less. Furthermore, the direction of this change was a function of whether or not a participant performed the specified behavior prior to the robot's cue. This was similarly observed both for frequency of behavior performance and for a second behavior (hands behind back). As such, this study informs future research on human-robot mimicry, particularly on the importance of prior behavior during a human-robot interaction. In doing so, this study provides a baseline for further understanding and exploring mimicry in human-robot interaction as well as evidence for a social component in human-robot mimicry.

This paper describes an extended (6-session) interaction between an ethnically and geographically diverse group of 26 first-grade children and the DragonBot robot in the context of learning about healthy food choices. We find that children demonstrate a high level of enjoyment when interacting with the robot, and a statistically significant increase in engagement with the system over the duration of the interaction. We also find evidence of relationship-building between the child and robot, and encouraging trends towards child learning. These results are promising for the use of socially assistive robotic technologies for long-term one-on-one educational interventions for younger children.

The development of trustworthy human-assistive robots is a challenge that goes beyond the traditional boundaries of engineering. Essential components of trustworthiness are safety, predictability and usefulness. In this paper we demonstrate that the integration of joint action understanding from human-human interaction into the human-robot context can significantly improve the success rate of robot-to-human object handover tasks. We take a two layer approach. The first layer handles the physical aspects of the handover. The robot's decision to release the object is informed by a Hidden Markov Model that estimates the state of the handover. Inspired by human-human handover observations, we then introduce a higher-level cognitive layer that models behaviour characteristic for a human user in a handover situation. In particular, we focus on the inclusion of eye gaze / head orientation into the robot's decision making. Our results demonstrate that by integrating these non-verbal cues the success rate of robot-to-human handovers can be significantly improved, resulting in a more robust and therefore safer system.