Planning handovers involving humans and robots in constrained environment

Abstract

Exchanging objects with humans through handovers is a key feature for any robot operating side by side with humans. The studies on the topic tackle various problems, such as the hand dexterity, the communication cues, the arms motions, the forces, the head motions, but most of them consider a handover as an independent action, decorrelated from the plan it is part of. In this paper, we consider situations where it might be necessary (or preferable) to achieve several handovers in order to transfer an object from one agent to another one. This problem complexity grows accordingly with the number of agents that might be involved in the task, making a classical approach under efficient. We propose a graph-based approach enabling a fast computation of a solution taking into account different parameters linked to the humans comfort and preferences. An abstract model of the task is also used as a heuristic to guide the search in the graph, a search which is performed with a Lazy Weighted A*. The method computes which agents (human or robot) to use and where handovers should be performed. It also computes motion plans for each robot, ensures that humans can reach handover places and preserves comfort of human partners by reducing, for instance, their efforts. The method has been implemented and is demonstrated on various simulated multi-agents environments and on our two PR2 robots interacting with two humans.

Benchmark for Human-to-Robot Handovers of Unseen Containers With Unknown Filling

“…The design of solutions for human-to-robot handovers can be informed by the behavior of human-to-human interactions in handovers through models of the relationship between the amount of force applied on the object by each party (grip force) and the proportion of the object's weight carried by each party (load force) [10], [26]- [28]. When modelling the motions between parties, solutions either find a complete trajectory (plan-based) [29]- [31] or continuously optimize the control output given the current state (controller-based) [32], [33]. However, even if controllers can generate natural motions, robots may stop moving right before transitioning to the passing phase, once both parties make contact with the object.…”

HATEB-2: Reactive Planning and Decision making in Human-Robot Co-navigation

“…Our previous work [4] is specifically developed to handle such intricate situations in semi-crowded environments. Such planning for humans along with the robot is usually required in robot-human handover scenarios, to know where to perform a task, and who performs a task [17], [18]. Similarly, in HATEB, the tightness of the elastic band can be adjusted to make either robot or the human take more load.…”

Viewing Robot Navigation in Human Environment as a Cooperative Activity

“…It provides additional advantage of streamlining robot behavior from open spaces to very constrained environmental conditions. In its spirit, our approach is similar to the previously proposed approaches for geometric [3] and symbolic [4] planning systems, where the robot synthesizes a shared plan for the human and itself.…”