Failure recovery with shared autonomy

This work assesses an adaptive approach to fault recovery in autonomous robotic space operations, which uses indicators of opportunity, such as physiological state measurements and observations of past human assistant performance, to inform future selections. We validated our reinforcement learning approach using data we collected from humans executing simulated mission scenarios. We present a method of structuring humanfactors experiments that permits collection of relevant indicator of opportunity and assigned assistance task performance data, as well as evaluation of our adaptive approach, without requiring large numbers of test subjects. Application of our reinforcement learning algorithm to our experimental data shows that our adaptive assistant selection approach can achieve lower cumulative regret compared to existing non-adaptive baseline approaches when using real human data. Our work has applications beyond space robotics to any application where autonomy failures may occur that require external intervention.

show abstract

“…It is recognized by the task monitor, a part of the onboard embedded agent. [5], [6], [7]. For example, Fig.…”

Section: Background and Problem Statementmentioning

confidence: 99%

Everybody Needs Somebody Sometimes: Validation of Adaptive Recovery in Robotic Space Operations

McGuire

Furlong

Fong

et al. 2019

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

show abstract

“…High-level goals are decomposed into subtasks by the planner to be completed by the executive. A task monitor observes the progress that the executive is making towards completion of each task [6], [7], [8]. Should the task either fail completely or be predicted to not meet requirements, an assistant selector is consulted to determine which assistant to consult.…”

Section: Background and Problem Statementmentioning

confidence: 99%

“…The subtask decomposition of higher-level goals is provided externally, e.g. by a dynamic task decomposition algorithms [22] or human intervention [6]. The subtask capabilities of the robot (t s v ) are fixed at deployment time.…”

Section: B Formal Assistance Allocation Problem Statementmentioning

confidence: 99%

Failure is Not an Option: Policy Learning for Adaptive Recovery in Space Operations

McGuire

Furlong

Heckman

et al. 2018

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

This paper considers the problem of how robots in long-term space operations can learn to choose appropriate sources of assistance to recover from failures. Current assistant selection methods for failure handling are based on manually specified static look up tables or policies, which are not responsive to dynamic environments or uncertainty in human performance. We describe a novel and highly flexible learningbased assistant selection framework that uses contextual multiarm bandit algorithms. The contextual bandits exploit information from observed environment and assistant performance variables to efficiently learn selection policies under a wide set of uncertain operating conditions and unknown/dynamically constrained assistant capabilities. Proof of concept simulations of long-term human-robot interactions for space exploration are used to compare the performance of the contextual bandit against other state of the art assistant selection approaches. The contextual bandit outperforms conventional static policies and non-contextual learning approaches, and also demonstrates favorable robustness and scaling properties.

show abstract

“…In this new class of intelligent automation systems, where automation systems include autonomous machines, long-term robustness without sacrificing flexibility will be a chal-lenging task. If the semi-autonomous robots of an industrial automation system can ask for help in intuitive ways [15], the existence of the human operators in the system becomes an enabler for intelligent automation systems, as they can intervene when something goes wrong [16]. This means that not only robustness is required, but also that it needs to be possible for an operator to influence the automation system to a large degree.…”

Section: Introductionmentioning

confidence: 99%

Application of the Sequence Planner Control Framework to an Intelligent Automation System with a Focus on Error Handling

2021

View full text Add to dashboard Cite

Future automation systems are likely to include devices with a varying degree of autonomy, as well as advanced algorithms for perception and control. Human operators will be expected to work side by side with both collaborative robots performing assembly tasks and roaming robots that handle material transport. To maintain the flexibility provided by human operators when introducing such robots, these autonomous robots need to be intelligently coordinated, i.e., they need to be supported by an intelligent automation system. One challenge in developing intelligent automation systems is handling the large amount of possible error situations that can arise due to the volatile and sometimes unpredictable nature of the environment. Sequence Planner is a control framework that supports the development of intelligent automation systems. This paper describes Sequence Planner and tests its ability to handle errors that arise during execution of an intelligent automation system. An automation system, developed using Sequence Planner, is subjected to a number of scenarios where errors occur. The error scenarios and experimental results are presented along with a discussion of the experience gained in trying to achieve robust intelligent automation.

show abstract

Failure recovery with shared autonomy

Cited by 10 publications

References 12 publications

Everybody Needs Somebody Sometimes: Validation of Adaptive Recovery in Robotic Space Operations

Everybody Needs Somebody Sometimes: Validation of Adaptive Recovery in Robotic Space Operations

Failure is Not an Option: Policy Learning for Adaptive Recovery in Space Operations

Application of the Sequence Planner Control Framework to an Intelligent Automation System with a Focus on Error Handling

Contact Info

Product

Resources

About