Intro overview

50 %
50 %
Information about Intro overview
Education

Published on January 16, 2008

Author: Siro

Source: authorstream.com

Autonomous Agents - Example Applications :  Autonomous Agents - Example Applications CS 569 Steve Chien Randall Hill steve.chien@jpl.nasa.gov http://www-aig.jpl.nasa.gov/ Voice: +1 (818) 393-5320 FAX: +1 (818) 393-5244 Portions of the research described in this paper was performed at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration. Outline:  Outline Metrics for Agent-hood Spacecraft Agents Rover Agents Game-playing Agents Ground Communications Station Agents Agent Characteristics:  Agent Characteristics Behave Intelligently Interact in an environment Exist over a persistent interval Accept Goals Often generate own goals Agents:  Agents Rovers Teams/Agents Antennas Spacecraft Computer Games UAVs Value of Autonomy:  Value of Autonomy Ensure robust operation after faults Reduce spacecraft operations costs Take advantage of science opportunities made possible by faster decision making Reduce spacecraft development costs Increased mission returns Innovations and Benefits:  Innovations and Benefits Goal-based commanding Expands high-level goals into low-level commands Expected to reduce operations costs Model-based reasoning Reasons using models rather than static responses Can react autonomously to unexpected situations Enables autonomous missions. Fail-operational fault recovery Tries alternate ways to achieve goals rather than simply safing the spacecraft. Reduces spacecraft “downtime” due to anomalies. Enables autonomous missions Slide8:  Remote Agent (RA) Architecture PS generates plan Exec expands & executes it MIR looks for failures. Exec can respond to fault locally... … or request a new plan. Planner: Diversity of goals:  Planner: Diversity of goals Final state goals “Turn off the camera once you are done using it” Scheduled goals “Communicate to Earth at pre-specified times” Periodic goals “Take asteroid pictures for navigation every 2 days for 2 hours” Information-seeking goals “Ask the on-board navigation system for the thrusting profile ” Continuous accumulation goals “Accumulate thrust with a 90% duty cycle” Default goals “When you have nothing else to do, point HGA to Earth” Planner: Diversity of constraints:  Planner: Diversity of constraints State/action constraints “To take a picture,the camera must be on.” Continuous time Finite resources Power True parallelism The ACS loops must work in parallel with the IPS controller Functional dependencies “The duration of a turn depends on its source and destination.” Continuously varying parameters Amount of accumulated thrust Other software modules as specialized planners On-board navigator Planning the first day:  Planning the first day 19 hours 5 hours Smart Executive:  Smart Executive Execution Support Language Multi-threaded operating system High-level control language Plan execution Plan request generation Activity decomposition Time driven and event driven execution Configuration management services Fault Protection Interface with MIR System-level fault protection Executing the first plan:  Executing the first plan MIR:  MIR Commands Monitored values State updates Goals Model Reconfiguration commands Model-based diagnosis in MIR:  Model-based diagnosis in MIR    Replanning the first day:  Replanning the first day Planning the second day:  Planning the second day Lander:  Lander DS4 Activities:  DS4 Activities Three sets of sample activities, each consisting of: Move the drill to the hole Drill the hole (mining) Move sample to oven and deposit Perform oven experiment (bake sample, take data) Let oven cool down before re-use There are also picture taking activities followed by data compression activities Engineering activities include: uplink from lander to spacecraft compress data in buffer Resources/States:  Resources/States Comm system Data Buffer Battery charge level Power Level Drill Camera (CIVA) Oven (2) state * Drill Location Camera state * Comm state * * can be failed Replan: Oven 1 Failure:  Replan: Oven 1 Failure Problem: Oven 1 has a complete failure. The 2nd and 3rd sample activities (as planned) use oven 1 Resolution: The planner replans the sample activities to use oven 2 General Repair Strategy: Conflict - state variable conflict Repair - substitute use of another resource (oven) to achieve sample activity planning system could also delete activity in conflict, this would require optimization to represent preference to have 3rd sample activity Replan: Oversubscribed Data Buffer:  Replan: Oversubscribed Data Buffer Problem: The actual data collected by the first sample activity is greater than expected (e.g. due to content dependent compression). The planner realizes that it will not have sufficient buffer memory to perform the third sample activity. Resolution: The planner adds an uplink activity after the second sample activity empties the data buffer. General Repair Strategy: Conflict - over-subscription of a depletable resource Repair - if there is a renewable activity (uplink) consider addition of such an activity Replan: Battery is depleted:  Replan: Battery is depleted Problem: The battery level is reduced in the simulator to a very low level. There is not enough battery charge to perform the third sample activity. Resolution: The simulator sends the new battery level to the planner. The planner realizes there is not enough power for the 3rd sample activity and deletes it. General Repair Strategy: Conflict - over-subscription of a depletable resource (battery charge) Repair - delete resource using activity Slide28:  Comet Lander Slide29:  Europa Submersible Slide31:  Current Strategy ... Alternate Strategy-1 Planner uses current strategy Other strategies proposed Strategies statistically evaluated based on performance in environment Utility “Best” strategy selected Process continues, enabling adaptation in a constantly changing environment Rover Agents:  Rover Agents Automated Onboard Planning of Rover Commands:  Automated Onboard Planning of Rover Commands ORCAA Input science requests Rover plan generation; execution monitoring; re-planning as necessary due to execution feedback Onboard s/w CASPER WITS ARCHITECTURE Architecture Components:  Architecture Components Planning: dynamically produces distributed rover operation plans Data Analysis: models rock type distribution and generates new science goals Rover Control S/W: handles low-level execution - used to control the JPL Rocky 7 rover Multi-Rover Hardware Simulator: simulates rover h/w operations within a terrain Environment Simulator: simulates different geological environments and manages science data Slide35:  Distributed Agents, Coordination & Teamwork System Diagram 9/98:  System Diagram 9/98 Science Planner Rover Science strategy portfolio distributed clustering geologic hypotheses rock distribution map geologic morphology map Environment Simulator rock, mineral spectra models rock texture models interface to navigation landscape geology location of each rover desired action (plan) actual new state rover sensor data prioritized list of desired observations Java Synchronization Client Java Synchronization Client asynchronous communicates over network distributed planning constraint tracking resource/state analysis conflict resolution heuristic search replanning rover sensor data prioritized list of desired observations desired action actual new state rover sensor data Aspen based target priorities Simulated Rockscapes Color coding: PCA on spectra:  Simulated Rockscapes Color coding: PCA on spectra Multiple Rovers in Simulated Rockscape:  Multiple Rovers in Simulated Rockscape Rock/Mineral Spectra Models:  Rock/Mineral Spectra Models Next: use Mars hydrothermal system-specific rock models from In Situ Data Analysis task. Distributable Clustering:  Distributable Clustering Spectra feature space (2 of 12 dimensions) Slide41:  Target Prioritization by Clustering geographic location of spectra sample cluster membership is indicated by color broad data collection goal - collect observation in this region specific data collection goal - collect observation at this location Slide44:  Market-Based Peer-to-Peer Approaches Computer Game Players:  Computer Game Players Real-time Strategy Games (e.g. Age of Kings (Microsoft Games/Ensemble Studios), … Building Competitive AI/Computer Players is challenging Background:  Background Play requires balance of Economy players must acquire resources (food, wood, gold, stone) players must advance technologically (by researching using resources) Military end goal is world conquest must build army, navy to defeat other players Team games (coordination, cooperation among team) Difficulties:  Difficulties Good players must have an overall strategy focus on economy first, then overpower enemy rush opponent to disrupt his economy deny resources to opponent, win war of attrition Combined arms tactics important Infantry, archers, seige weapons all important Good Computer Player Must:  Good Computer Player Must Be able to chose a viable strategy Manage/Balance resource acquisition for economy Respond to exogenous events (enemy attacks) Manage Assaults on opposing players Acquire intelligence on opposing players DSN Automation:  Ground Communication Station Automation DSN Automation Deep-Space Terminal (DS-T) Automation:  Deep-Space Terminal (DS-T) Automation Goal: enable a goal-based interface to the station tell the station what you want achieved, not how to do it Benefit: enable significant cost reduction by automating the process of capturing spacecraft data. DS-T is a prototype for autonomous ground station operation. Deep Space Terminal (DS-T):  Deep Space Terminal (DS-T) What Goals specified through a high level Service Request HOW DS-T figures out all the steps necessary STOL scripts are: selected sequenced parameterized Station verifies correct execution responding to anomalies as required DS-T, 34m Antenna DSS-26 Goldstone, CA:  DS-T, 34m Antenna DSS-26 Goldstone, CA DS-T:  DS-T What is Track Planning? Automated Proced ure Generation of DSN communication antenna command sequences Configure_equipment: Start jsc_asn.prc(dss,sc,pass,&ret_status) If (!ret_status) then Write(“fatal error: cannot start pass”) Goto fatal_err Endif Start ugc_hi.prc If (!ret_status) then Write(“fatal error: can’t control UGC”) Goto fatal_err Endif Start apc_hi.prc If (!ret_status) then Write(“fatal error: can’t control APC”) Goto fatal_err Endif ... Point_antenna: Ret_status = exec(“APC DCOS”) Start apc_track.prc(&ret_status) If (!ret_status) then Write(“fatal error: cannot point ant”) Goto fatal_err Endif ... Automated Track Plan Generation: The Problem:  Automated Track Plan Generation: The Problem Given: antenna tracking services requested by project (e.g. perform downlink track w/ MGS) equipment assignment (e.g. Block V receiver, APC antenna controller) Produce: an ordered set of actions (plan) to provide the requested communications link Why is Track Operations Hard?:  Why is Track Operations Hard? Many different combinations of services (> 100) Many different equipment configurations (> 400) must consider interactions (requires considerable antenna operations knowledge) interactions selection & sequencing parameterization some independence among how to achieve services 700 manual inputs under nominal conditions Input Examples:  Input Examples Goals/Info from track service request spacecraft (e.g. MGS, CAS, MPF) service (e.g. downlink, uplink) begin/end times for pre-track, track and post-track segments Other track information (from SOE) bit rate changes loss of signal occurrences way changes predict information carrier frequency, symbol rate, etc. Slide61:  Inputs and Outputs for Track Plan Generation DS-T Impact:  DS-T Impact Demonstrated the feasibility of “Lights-Out” operations for antenna ground stations. Automated operations of partial MGS pass (5/98) Automated operations of complete downlink passes MGS (May - August 1998 ) “Lights-Out” operations 6 day demonstration DS-T automated MGS downlink for 6 continuous days (August 1998) Achieved comparable to human performance (90% FR) DS-T is the baseline for the Network Simplification Plan (NSP) Benefits of Automated Track Plan Generation:  Benefits of Automated Track Plan Generation Enabled autonomous operations by eliminating need for manual inputs Reduced level of required expertise for an operator to perform a communications track Antenna KB developed for this task provides a declarative representation of antenna operations procedures Summary:  Summary Intelligent Agents are applicable to a wide range of problems AI techniques are applicable to many Agent problems Many high impact and high profile problems exist Many research problems are still outstanding

Add a comment

Related presentations

Related pages

Basecamp 3 - Intro & Overview - YouTube

Introducing Basecamp 3! ... Show #38 - Doing onboarding/intro videos and the new client side feature in Basecamp - Duration: 29:26.
Read more

Overview – Intro - State of the News Media 2015

Overview – Intro . Introduction. By the Project for Excellence in Journalism. Some of the numbers are chilling. Newspaper ad revenues have fallen 23% in ...
Read more

Link - AVAILABLE NOW! - Intro & Overview - YouTube

Link - Intro & Overview Wooo! It's finally complete! After tons of hours smashing my head on my keyboard now you don't have to! Introducing Link ...
Read more

Intro Videos Overview - code.visualstudio.com

Start your journey with Visual Studio Code with this set of introductory videos. The videos are designed to give you a quick overview of the ...
Read more

Intro and overview - ASM International

Support. We provide a full range of service and support options, delivered world-wide through ASM’s global technical and logistics infrastructure by ...
Read more

Intro Overview – DDT Pro Training

Intro Overview May 25, 2016 Don Keller. https://ddtprotraining.com/videos/DDTProOverview.mp4
Read more

Vegas Product Family Overview - Sony Creative Software

Sony Creative Software is pleased to announce that MAGIX Software GmbH has purchased the majority of SCS software products and other assets.
Read more

Flash Intro Templates | Flash Intros - Template Monster

Flash Intro Templates are made to jazz up your e-space, create compelling motion graphics to tell about your business, represent website or product.
Read more

Introduction to SciFinder Tutorial

Introduction to SciFinder Tutorial
Read more