Shared autonomy via hindsight optimization for teleoperation and teaming

By Shervin Javdani, Henny Admoni, Stefania Pellegrinelli, Siddhartha S. Srinivasa, J. Bagnell, Shervin Javdani Carnegie. 2017

260 citations recorded by the source corpus.

In shared autonomy, a user and autonomous system work together to achieve shared goals. To collaborate effectively, the autonomous system must know the user’s goal. As such, most prior works follow a predict-then-act model, first predicting the user’s goal with high confidence, then assisting given that goal. Unfortunately, confidently predicting the user’s goal may not be possible until they have nearly achieved it, causing predict-then-act methods to provide little assistance. However, the system can often provide useful assistance even when confidence for any single goal is low (e.g. move towards multiple goals). In this work, we formalize this insight by modeling shared autonomy as a partially observable Markov decision process (POMDP), providing assistance that minimizes the expected cost-to-go with an unknown goal. As solving this POMDP optimally is intractable, we use hindsight optimization to approximate. We apply our framework to both shared-control teleoperation and human–robot teaming. Compared with predict-then-act methods, our method achieves goals faster, requires less user input, decreases user idling time, and results in fewer user–robot collisions.

Subfields: Cognitive Neuroscience, Mechanical Engineering, Reinforcement Learning, Human-Computer Interaction

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Neuroscience

In Mechanical Engineering

In Reinforcement Learning

In Human-Computer Interaction

Citations in the corpus

LILA: Language-Informed Latent Actions (2021, background): “Shared autonomy casts robot control as a collaborative process between humans and robots [2, 4, 16, 22].”
Shared Autonomy with Learned Latent Actions (2020, methodology): “Similar to [8] and [16], let the robot provide assistance towards each discrete goal g ∈ G in proportion to the robot’s confidence in that goal1:”
Learning latent actions to control assistive robots (2021, unknown): “Shared autonomy approaches like [16,37,25] are not suitable within this setting because: i) the goals lie in continuous regions and ii) the user needs to control both the high-level goal that the robot reaches for and the trajectory the robot follows to reach that goal (e.”
HARMONIC: A multimodal dataset of assistive human–robot collaboration (2018, methodology): “Finally, for the original robot control code, follow this Prepared using sagej.cls link to a fork of the publicly available implementation of the shared autonomy code we used: https: //github.com/HARPLab/ada_meal_scenario (Javdani et al. 2015).”
Learning User-Preferred Mappings for Intuitive Robot Control (2020, methodology): “is to use control interfaces such as joysticks or sip-andpuff devices [6], [7].”
Controlling Assistive Robots with Learned Latent Actions (2019, methodology): “with two 2-DoF joysticks, and were given a button to toggle between linear and angular motion [3], [7], [25].”
Probabilistic Human Intent Recognition for Shared Autonomy in Assistive Robotics (2019, methodology): “1 (RBII-1 and RBII-2), for comparative purposes we also implement two approaches to intent inference utilized in previous shared autonomy works—amnesic inference [2, 9, 13] and memory-based inference [8, 9, 20, 30].”