It Takes Two: Learning to Plan for Human-Robot Cooperative Carrying

Cooperative table-carrying is a complex task due to the continuous nature of the action and state-spaces, multimodality of strategies, and the need for instantaneous adaptation to other agents. In this work, we present a method for predicting realistic motion plans for cooperative human-robot teams on the task. Using a Variational Recurrent Neural Network (VRNN) to model the variation in the trajectory of a human-robot team across time, we are able to capture the distribution over the team's future states while leveraging information from interaction history. The key to our approach is leveraging human demonstration data to generate trajectories that synergize well with humans during test time in a receding horizon fashion. Comparison between a baseline, sampling-based planner RRT (Rapidly-exploring Random Trees) and the VRNN planner in centralized planning shows that the VRNN generates motion more similar to the distribution of human-human demonstrations than the RRT. Results in a human-in-the-loop user study show that the VRNN planner outperforms decentralized RRT on task-related metrics, and is significantly more likely to be perceived as human than the RRT planner. Finally, we demonstrate the VRNN planner on a real robot paired with a human teleoperating another robot.

I. Introduction

Humans internally develop and rely on models of the world around them to make goal-oriented decisions. They have internalized strategies that drive their decision-making based on their goals and observations. In collaborative tasks, human strategies must accommodate their teammates' behaviors in order to successfully meet their mutual objectives. Robots that collaborate with humans must emulate this ability to be effective teammates [1], [2]. Achieving this capability is especially difficult due to the inherently multimodal nature of human interactions. However, even when robots are equipped with an underlying prior on human actions, human actions are extremely noisy and difficult to predict. This situation is especially true in complex, continuous action and state space tasks. Learning the sampling distributions over future motion sequences from successful human demonstrations not only allows the robot to plan over potential outcomes, but also come up with viable actions that result in more fluent interactions (e.g. not pushing the table towards each other, or pulling in opposite directions).