Learning Robot Objectives from Physical Human Interaction

By Andrea V. Bajcsy, Dylan P. Losey, Marcia Kilchenman O'Malley, Anca Dragan. 2017

167 citations recorded by the source corpus.

Subfields: Reinforcement Learning, Human-Computer Interaction, Robotics & Control

How researchers see this paper

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

In Reinforcement Learning

In Human-Computer Interaction

In Robotics & Control

Citations in the corpus

Learning from Physical Human Corrections, One Feature at a Time (2018, methodology): “Following [3], we interpret the change in configuration ∆qt as an indication of the corrected trajectory, ξ t c , that the human would prefer for the robot to execute:”
Recent Advances in Robot Learning from Demonstration (2020, methodology): “To circumvent this issue, LfD has been successfully applied to trajectory optimization-based methods by assuming that the expert minimizes a hidden cost function when demonstrating a skill (96, 110, 111, 112).”
Learning from Interventions: Human-robot interaction as both explicit and implicit feedback (2020, background): “Similar ideas have been successfully applied to reduce effort in learning from kinesthetic corrections in the manipulator control domain [23, 24, 25] or by monitoring a buffer of future robot actions in social robotic teaching [26].”
Correcting Robot Plans with Natural Language Feedback (2022, background): “This problem has been also addressed by learning to predict true rewards given observed rewards across environments and tasks [16], and by learning from physical interactions with humans [5].”
Learning Interactively to Resolve Ambiguity in Reference Frame Selection (2020, unknown): “[2, 3, 4].”
Reward-rational (implicit) choice: A unifying formalism for reward learning (2020, background): “[11] propose that corrections are informative about the reward and use the propagation as their grounding, deriving an approximate MAP estimate for the reward.”
Feature Expansive Reward Learning: Rethinking Human Input (2020, methodology): “Recent methods interpret such corrections as evidence about the human’s desired preference for how to complete the task, enabling the robot to update its reward function online [1–3].”
Consequences of Misaligned AI (2021, background): “This includes systems that learn from direct comparisons between options (9; 11; 13), demonstrations of optimal behavior (26; 1; 36), corrective feedback (7; 15; 2), and proxy metrics (18).”