Demonstration-based control of supernumerary robotic limbs

Abstract

The body representation in the human mind is dynamic, and illusions or traumatic events can modify it to include additional limbs. This remarkable adaptability of the central nervous system to different body configurations opens new possibilities in the field of human augmentation. In order to fully exploit this potential, we developed a new type of wearable co-robot that can perform tasks in close coordination with the human user. The system, named Supernumerary Robotic Limbs (SRL), consists of two additional robotic arms worn through a backpack-like harness. The SRL can assist the user by holding objects, lifting weights and streamlining the execution of a task. If the SRL perform movements closely coordinated with the user and exhibit human-like dynamics, they might be incorporated into the body representation and perceived as parts of the user's body. As a result, the human would be able to extend the range of available skills and manipulation possibilities, performing tasks more effectively and with less effort. This paper presents a communication, estimation and control method for the SRL, aimed to perform tasks in tight coordination with the wearer. The SRL observes the user motion, and actively assists the human by employing a coordinated control algorithm. In particular, skills involving the direct cooperation of two human workers are transferred to the SRL and a single user. Demonstration data of the two humans - a leader and an assistant - are analyzed and a state estimation algorithm is extracted from them. This can be used to control the SRL accordingly with the used end effectors. A causal relationship relating the assistant's motion to the leader's motion is identified based on System Identification methods. This approach is applied to a drilling operation performed by two workers. An effective coordination skill is identified and transferred to the SRL, to make them act like the human follower.