Online Hybrid Motion Planning for Dyadic Collaborative Manipulation via Bilevel Optimization

Theodoros Stouraitis,Sethu Vijayakumar,Iordanis Chatzinikolaidis,Michael Gienger

doi:10.1109/tro.2020.2992987

Theodoros Stouraitis, Sethu Vijayakumar + Show 2 more

Open Access

https://doi.org/10.1109/tro.2020.2992987

Copy DOI

Abstract

Effective collaboration is based on online adaptation of one's own actions to the actions of their partner. This article provides a principled formalism to address online adaptation in joint planning problems such as Dyadic collaborative Manipulation (DcM) scenarios. We propose an efficient bilevel formulation that combines graph search methods with trajectory optimization, enabling robotic agents to adapt their policy on-the-fly in accordance to changes of the dyadic task. This method is the first to empower agents with the ability to plan online in hybrid spaces; optimizing over discrete contact locations, contact sequence patterns, continuous trajectories, and force profiles for co-manipulation tasks. This is particularly important in large object co-manipulation that requires changes of grasp-holds and plan adaptation. We demonstrate in simulation and with robot experiments the efficacy of the bilevel optimization by investigating the effect of robot policy changes in response to real-time alterations of the dyadic goals, eminent grasp switches, as well as optimal dyadic interactions to realize the joint task.

Highlights

We introduce a decomposition of general hybrid motion into a set of hybrid motion primitives, referred as the Hybrid Optimization Lexicon for Manipulation (HOLM)
3) Discussion: The main steps that allow us to improve the computation times from tens of seconds in our previous work [7], to milliseconds for HOLM and few seconds for the bilevel optimization are: (i) Decomposing the problem into HOLM primitives, which allows to keep the size of the hybrid problems small, (ii) exploring the hybrid structure of the problem with an efficient graph-search algorithm, (iii) formulating a sparse problem that can be efficiently solved6, (iv) providing the exact Hessian using automatic differentiation, and (v) selecting the end-effectors’ and permissible force representation discussed in Sections VI-3 and VI-4
This article presents a novel concept towards online adaptive robot motion generation for physical human-robot collaboration tasks, such as Dyadic collaborative Manipulation (DcM) scenarios

Summary

Introduction

The two individuals partner together to form a distributed system, augmenting their manipulation abilities Such individuals can be either humans or robots. The policy of the agent πa generates hybrid action trajectories, that guide the object from the current state to the goal. We illustrate one such trajectory, where the object’s pose yt, the end-effectors’ positions ck and the contact force f l of the left end-effector are visualized. Such trajectories have hybrid nature due to the contact change. In manipulation setups the structure of the motion specifies the arms’ contact sequence pattern, i.e. the order with which the arms change contacts

Objectives

Discussion

Conclusion