Intent-Recognition-Based Traded Control for Telerobotic Assembly over High-Latency Telemetry
Johns Hopkins University
As we deploy robotic manipulation systems into unstructured real-world environments, the tasks which those robots are expected to perform grow very quickly in complexity. These tasks require a greater number of possible actions, more variable environmental conditions, and larger varieties of objects and materials which need to be manipulated. This in turn leads to a greater number of ways in which elements of a task can fail. When the cost of task failure is high, such as in the case of surgery or on-orbit robotic interventions, effective and efficient task recovery is essential. Despite ever-advancing capabilities, however, the current and near future state-of-the-art in fully autonomous robotic manipulation is still insufficient for many tasks in these critical applications. Thus, successful application of robotic manipulation in many application domains still necessitates a human operator to directly teleoperate the robots over some communications infrastructure. However, any such infrastructure always incurs some unavoidable round-trip telemetry latency depending on the distances involved and the type of remote environment. While direct teleoperation is appropriate when a human operator is physically close to the robots being controlled, there are still many applications in which such proximity is infeasible. In applications which require a robot to be far from its human operator, this latency can approach the speed of the relevant task dynamics, and performing the task with direct telemanipulation can become increasingly difficult, if not impossible. For example, round-trip delays for ground-controlled on-orbit robotic manipulation can reach multiple seconds depending on the infrastructure used and the location of the remote robot. The goal of this thesis is to advance the state-of-the art in semi-autonomous telemanipulation under multi-second round-trip communications latency between a human operator and remote robot in order to enable more telerobotic applications. We propose a new intent-recognition-based traded control (IRTC) approach which automatically infers operator intent and executes task elements which the human operator would otherwise be unable to perform. What makes our approach more powerful than the current approaches is that we prioritize preserving the operator's direct manual interaction with the remote environment while only trading control over to an autonomous subsystem when the operator-local intent recognition system automatically determines what the operator is trying to accomplish. This enables operators to perform unstructured and a priori unplanned actions in order to quickly recover from critical task failures. Furthermore, this thesis also describes a methodology for introducing and improving semi-autonomous control in critical applications. Specifically, this thesis reports (1) the demonstration of a prototype system for IRTC-based grasp assistance in the context of transatlantic telemetry delays, (2) the development of a systems framework for IRTC in semi-autonomous telemanipulation, and (3) an evaluation of the usability and efficacy of that framework with an increasingly complex assembly task. The results from our human subjects experiments show that, when incorporated with sufficient lower-level capabilities, IRTC is a promising approach to extend the reach and capabilities of on-orbit telerobotics and future in-space operations.
robotics, human-robot interaction, space, space robots, teleoperation, telemanipulation, semi-autonomy, supervised control, intent recognition, human factors, on-orbit assembly, robotic assembly, satellite servicing