MBD in realworld system SelfConfiguring Systems Meir Kalech
MBD in real-world system… Self-Configuring Systems Meir Kalech Partially based on slides of Brian Williams
Outline p p Last lecture: 1. Models of correct + faulty behavior 2. Sherlock engine 3. Abductive diagnosis 4. Qualitative models Today’s lecture: 1. Autonomous systems 2. Model-based programming 3. Livingstone
Motivation p Machines are increasingly aware of themselves & environment p They are increasingly able to detect and respond to conditions p What is the next level of awareness, robustness, adaptivity?
NASA Research Challenges p p p Some machines must survive years without repair Relatively short down time can destroy a mission Development & operation costs must be contained Challenge: Easily developed, highly capable control systems
Problem Statement p Given n n A model of a physical system such as a spacecraft The internal actions taken and observations Configuration Goals State Estimate Model Observations p Action Selection Command Determine n n The most likely internal states of the system The commands needed to move to a desirable state
Typical Domain p Engineers model the local, qualitative behavior of system components n Components are things like valves, switches, tanks, engines n Properties of interest are transmission of flow, voltage, etc n Goals are “produce acceleration”, “maintain pointing ability”, etc
Spacecraft Engine System Model main engines Fuel tank Regulators Pyro valves Helium tank Acc latch valves oxidizer tank • Helium pressurizes the fuel and oxidizer tanks with the regulators which control the high pressure. • Acc senses the thrust generated by the engines.
main engines latch valves Fuel tank Acc Regulators Pyro valves Helium tank Spacecraft Engine System Model oxidizer tank High level goal: producing thrust Several configurations: 1. Open latch valves in the left engine. 2. Firing pyro valves and open a set of latch valves to the right engine. 3. More configurations of valves states…
main engines latch valves Fuel tank Acc Regulators Pyro valves Helium tank Spacecraft Engine System Model oxidizer tank • Suppose configuration 1 is selected. • Configuration 1 failed – not enough thrust. • Find lowest cost new configuration that satisfies goals.
Outline p p Last lecture: 1. Models of correct + faulty behavior 2. Sherlock engine 3. Abductive diagnosis 4. Qualitative models Today’s lecture: 1. Autonomous systems 2. Model-based programming 3. Livingstone
Model-based Program Evolves Hidden State Programmer specifies abstract state evolutions Thrust Goals Programmer specifies plant model Delta_V(direction=b, magnitude=200) Power Temporal planner Attitude Point(a) Engine Off Model specifies • Mode transitions • Mode behavior State Valve 0. 01 Open Model 0. 01 Open goals Model-based Executive Stuck open Close 0. 01 Closed Off Flight System Control Observations. Stuck 0. 01 closed inflow = outflow = 0 Control Layer Command
Model-based Executive Reasons from Plant Model Goal: Achieve Thrust Goals Delta_V(direction=b, magnitude=200) Power State Estimates State Goals Temporal Point(a) Attitude Estimate & Diagnose planner Off Reconfigure & Repair Engine Off State Estimates State Goals Off Observations Engine Commands Model-based Executive Open four valves Model Observations Flight System Control Layer Commands
Model-based Executive Reasons from Plant Model Goal: Achieve Thrust Goals Delta_V(direction=b, magnitude=200) Power State Estimates State Goals Temporal Attitude Estimate & Diagnose planner Point(a) Off Reconfigure & Repair Engine State Off goals Diagnose: Valve fails stuck closed Model-based Executive Model Observations Flight System Control Layer Command Switch to backup
Outline p p Last lecture: 1. Models of correct + faulty behavior 2. Sherlock engine 3. Abductive diagnosis 4. Qualitative models Today’s lecture: 1. Autonomous systems 2. Model-based programming 3. Livingstone
(Livingstone) commanded NASA’s Deep Space One probe Started: January 1996 Launch: October 15 th, 1998 Remote Agent Experiment: May, 1999 courtesy NASA JPL
Livingstone [Williams & Nayak, AAAI 96] State goals State estimate Model Mode Estimation Mode Reconfiguration Flight System Control Observations Control Layer Command
Estimate current likely Modes Reconfigure Thrustmodes to meet goals State estimate Model Mode Estimation Mode Selection Observations Flight System Control Command RT Control Layer
Mode Estimation: Mode Selection: Select a most likely set of component mode transitions that are consistent with the model and observations Select a least cost set of allowed component modes that entail the current goal, and are consistent State estimate Model arg max Pt(m’) Mode Estimation State goals Mode Selection arg min Ct(m’) s. t. M(m’) entails G(m’) s. t. M(m’) ^ O(m’) is consistent Observations Flight System Control Command s. t. M(m’) is consistent P – probability, M – modes, ORT Control Layer - observations C – cost, G - goals
Current Demonstration Testbeds p p p p Air Force Tech Sat 21 flight NASA NMP ST-7 Phase A NASA Mercury Messenger on ground. MIT Spheres on Space Station NASA Robonaut, X-37, ISPP Multi-Rover Testbed Simulated Air Vehicles
- Slides: 19