Download presentation
Presentation is loading. Please wait.
Published byAntwan Brafford Modified over 9 years ago
1
Thrust ID: Peer-to-Peer HRI Training and Learning with Humans Rod Grupen (lead) Cynthia Breazeal Nicholas Roy MURI 8 Kickoff Meeting 2007
2
MURI 8 Kickoff Meeting 2007 Interactive Training in Human-Robot Teams MIT-Vanderbilt-Stanford UW-UMASS Amherst Learning from human demonstration common ground - project proprietary sensory and motor policy spaces into a common frame asking questions and providing explanations combine learned policies flexibly and effectively in response to new run-time situations perform peer-to-peer policies jointly with humans and other robots
3
MURI 8 Kickoff Meeting 2007 Thrust ID Objectives MIT-Vanderbilt-Stanford UW-UMASS Amherst build libraries of schema for component tasks underlying triage, hazmat, and HRI acquaint human partners with capabilities and limitations of robot partners establish common knowledge about strategies, procedures, and practices negotiate roles and preferences for joint activity master joint activities through practice
4
MURI 8 Kickoff Meeting 2007 Actions MIT-Vanderbilt-Stanford UW-UMASS Amherst g g K g T g g’
5
MURI 8 Kickoff Meeting 2007 Developmental Programming MIT-Vanderbilt-Stanford UW-UMASS Amherst l native control basis l temporal structure for development l integration - hard problems use previous solutions for easier problems teleological hypothesis
6
MURI 8 Kickoff Meeting 2007 Action Schemas - Hierarchy MIT-Vanderbilt-Stanford UW-UMASS Amherst computational model of infant development stage 1 - touch what you see stage 2 - the length of your arm stage 3 - grasp affordances stage 4 - human collaboration Vgotskian pointing multi-body objects: simultaneous trackability attributes: motion (scale, multi-body kinematics), topological/geometrical attributes, hue, saturation, intensity, texture
7
MURI 8 Kickoff Meeting 2007 Action Schemas - Generative Models, Teleology, and Transfer Learning MIT-Vanderbilt-Stanford UW-UMASS Amherst teleoperator sorting instruction sorting replay with prior knowledge (1) parse events to find a matching schema. (2) associate goals with schema (3) Replicate demonstration with contingencies
8
MURI 8 Kickoff Meeting 2007 Commodity Mobile Manipulators MIT-Vanderbilt-Stanford UW-UMASS Amherst …nature routinely selects for dynamics to combine speed and agility with light weight and low power… strength, performance, safety
9
MURI 8 Kickoff Meeting 2007 Whole-Body Primate/Hominid/Human Model MIT-Vanderbilt-Stanford UW-UMASS Amherst postural stability prehensile skills tool use social organization
10
MURI 8 Kickoff Meeting 2007 Contributions MIT-Vanderbilt-Stanford UW-UMASS Amherst multi-agent, and human-robot schema for coordinated action interactive, socially-guided learning from demonstration (question/explain) hierarchical composition of skills communicative actions to convey states, objects, and actions
11
MURI 8 Kickoff Meeting 2007 Year 1 Demonstrations MIT-Vanderbilt-Stanford UW-UMASS Amherst component schema for initial triage client sideremote network client side remote network
12
MURI 8 Kickoff Meeting 2007 Year 1 Demonstrations MIT-Vanderbilt-Stanford UW-UMASS Amherst sample acquisition, and cataloging load carrying strategies that do not violate stability constraints of the platform tool use hazard containment
13
MURI 8 Kickoff Meeting 2007 Year 2 (and onward) MIT-Vanderbilt-Stanford UW-UMASS Amherst role engagement and switching in multi-robot, and human-robot strategies remote humans, prior knowledge, maps, run- time situational awareness, mental models, asymmetric beliefs, affect on communicative actions UGV/UAV/human coalitions
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.