Regions of Inaccurate Modeling for Robot Anomaly Detection and Model Correction

Speaker: Juan Pablo Mendoza , Carnegie Mellon University

Date: Monday, May 22, 2017

Time: 2:00 PM to 3:00 PM Note: all times are in the Eastern Time Zone

Public: Yes

Location: Seminar Room G882 (Hewlett Room)

Event Type:

Room Description:

Host: Brian Williams, Model-based Embedded & Robotics Systems Group, MIT

Contact: Christian Muise, 617-417-2068, cjmuise@csail.mit.edu

Relevant URL: http://www.cs.cmu.edu/~jmendoza/

Speaker URL: None

Speaker Photo:
None

Reminders to: seminars@csail.mit.edu, robotics@mit.edu

Reminder Subject: TALK: Regions of Inaccurate Modeling for Robot Anomaly Detection and Model Correction

Title: Regions of Inaccurate Modeling for Robot Anomaly Detection and Model Correction

Time: 2pm - 3pm, May 22, 2017
Location: Stata G882 (Hewlett Room)

Abstract:
To make intelligent decisions, robots often use models of the stochastic effects of their actions on the world. Unfortunately, in complex environments, it is often infeasible to create models that are accurate in every plausible situation, which can lead to suboptimal performance. We present an approach to enable robots to act robustly in the presence of model inaccuracies that are subtle --i.e., they cannot be detected from a single observation-- and context-dependent --i.e., they affect particular regions of the robot's state-action space. Our approach consists of enabling robots to explicitly reason about parametric Regions of Inaccurate Modeling (RIMs) in their state-action space. We enable robots to detect these RIMs from sparse execution data, to correct their models given these detections, and to plan accounting for uncertainty with respect to these RIMs. We demonstrate our work on the CMDragons team of soccer-playing robots, and on the CoBot motion service robot.

Bio:
Juan Pablo Mendoza is a PhD candidate in the Robotics Institute at Carnegie Mellon University. His research focuses on improving the robustness of autonomous robots at execution time via execution monitoring, anomaly detection and online learning. He has been the leader of the CMDragons team of soccer-playing robots in 2015 and 2016, with whom he has won the Small Size League of RoboCup in 2015, and obtained second place in 2013, 2014 and 2016.

Research Areas:

Impact Areas:

This event is not part of a series.

Created by Christian Muise at Tuesday, May 16, 2017 at 2:51 PM.