Multi-Level Learning in Half breed Deliberative/Responsive Portable Robot Structural Programming Frameworks

Multi level learning in hybrid deliberative reactive mobile robot architectural software systems l.jpg
1 / 21
1063 days ago, 339 views
PowerPoint PPT Presentation
Portable Intelligence Inc. Dr. Doug MacKenzie. Georgia Tech/Mobile ... Georgia Tech/Mobile Intelligence. 11. 5. Probabilistic Planning and Execution

Presentation Transcript

Slide 1

´╗┐Multi-Level Learning in Hybrid Deliberative/Reactive Mobile Robot Architectural Software Systems DARPA MARS Kickoff Meeting - July 1999

Slide 2

Georgia Tech College of Computing Prof. Ron Arkin Prof. Ashwin Ram Prof. Sven Koenig Georgia Tech Research Institute Dr. Tom Collins Mobile Intelligence Inc. Dr. Doug MacKenzie Personnel

Slide 3

Impact Provide the DoD people group with a stage free robot mission particular framework, with cutting edge learning capacities Maximize utility of automated resources in combat zone operations Demonstrate warfighter-arranged devices in three settings: reproduction, lab robots, and government-outfitted stages

Slide 4

New Ideas Add machine learning ability to a demonstrated robot-autonomous engineering with a client acknowledged human interface Simultaneously investigate five distinctive learning approaches at proper levels inside a similar design Quantify the execution of both the robot and the human interface in military-significant situations

Slide 5

Case-based Reasoning for: deliberative direction ("wizardry") receptive situational-subordinate behavioral setup Reinforcement learning for: run-time behavioral conformity behavioral collection choice Probabilistic behavioral moves gentler setting exchanging knowledge based arranging direction Adaptation and Learning Methods Available Robots and MissionLab Console

Slide 6

Reactive level engine blueprints behavioral combination by means of additions Deliberative level Plan encoded as FSA Route organizer accessible AuRA - A Hybrid Deliberative/Reactive Software Architecture

Slide 7

Reactive learning through element pick up modification (parametric alteration) Continuous adjustment in view of late experience Situational examinations required basically: If it works, continue doing it somewhat harder; in the event that it doesn't, take a stab at something else 1. Learning Momentum

Slide 8

Another type of responsive learning Previous frameworks include: ACBARR and SINS Discontinuous behavioral exchanging 2. CBR for Behavioral Selection

Slide 9

Reinforcement learning at coarse granularity (behavioral assem-blage choice) State space tractable Operates at level above learning force (choice rather than alteration) 3. Q-learning for Behavioral Assemblage Selection

Slide 10

Experience-driven help with mission particular At deliberative level above existing arrangement representation (FSA) Provides mission arranging support in setting 4. CBR "Wizardry"

Slide 11

"Gentler, kinder" technique for coordinating circumstances and their perceptual triggers Expectations created in light of situational probabilities in regards to behavioral execution (e.g., obstruction densities and navigability), utilizing them at arranging stages for behavioral determination Markov Decision Process, Dempster-Shafer, and Bayesian strategies to be examined 5. Probabilistic Planning and Execution

Slide 12

Integration with MissionLab Usability-tried Mission-detail programming created under DARPA financing (RTPC/UGV Demo II/TMR programs) Incorporates demonstrated and novel machine learning abilities Extends and implants deliberative Autonomous Robot Architecture (AuRA) capacities Architecture Subsystem Specification Mission Overlay

Slide 13

Development Process with Mlab Behavioral Specification MissionLab Simulation Robot

Slide 14

MissionLab Example: Scout Mission

Slide 15


Slide 16

MissionLab Example: Trashbot (AAAI Robot Competition)

Slide 17

MissionLab Reconnaissance Mission Developed by University of Texas at Arlington utilizing MissionLab as a major aspect of UGV Demo II Coordinated sensor indicating crosswise over arrangements

Slide 18

Evaluation: Simulation Studies Within MissionLab test system structure Design and choice of applicable execution criteria for MARS missions (e.g., survivability, mission consummation time, mission unwavering quality, cost) Potential expansion of DoD test systems, (e.g., JCATS)

Slide 19

Drawn from our current armada of versatile robots Annual Demonstrations Evaluation: Experimental Testbed

Slide 20

Evaluation: Formal Usability Studies Test in convenience lab Subject pool of competitor end-clients Used for both MissionLab and group teleautonomy Requires create ment of ease of use criteria and measurements

Slide 21