Lecture 14 Future Directionsmurray/courses/eeci-sp11/L14_directions-25Mar11.pdfNew tools for...

25
Lecture 14 Future Directions Richard M. Murray Caltech Control and Dynamical Systems 25 March 2011 Goals: Summarize main motivations and conclusions of the course Summarize (semi-) recent reports on future directions in control - 2001, 2005, 2010 snapshots Bigger picture: alternative architectures for control Reading: http://www.cds.caltech.edu/~murray/cdspanel http://www.cds.caltech.edu/~murray/topten http://paths.lids.mit.edu (offline?)

Transcript of Lecture 14 Future Directionsmurray/courses/eeci-sp11/L14_directions-25Mar11.pdfNew tools for...

Lecture 14 Future Directions

Richard M. MurrayCaltech Control and Dynamical Systems

25 March 2011Goals:

• Summarize main motivations and conclusions of the course• Summarize (semi-) recent reports on future directions in control

- 2001, 2005, 2010 snapshots• Bigger picture: alternative architectures for control

Reading:• http://www.cds.caltech.edu/~murray/cdspanel• http://www.cds.caltech.edu/~murray/topten• http://paths.lids.mit.edu (offline?)

Richard M. Murray, Caltech CDSEECI, Mar 2011

OnlineOptimization(RHC, MILP)

Sensing

External Environment

ProcessActuation Feeder:Reliable

StateServer

(KF -> MHE)

Inner Loop(PID, H∞)

Com

man

d:FI

FO

1-3 Gb/s

Sensing

Traj:Causal

ActuatorState:Unreliable

Map:CausalTraj:Causal

10 Mb/s

Goal Mgmt(MDS)

Attention &Awareness

Memory andLearning

“Controls” 2010: Networked Systems(following P. R. Kumar)

Online Model

StateServer

(KF -> MHE)

State:Unreliable

StateServer

(KF, MHE)

100 Kb/s

OnlineOptimization(RHC, MILP)

OnlineOptimization(RHC, MILP)

Mode andFault

Management

¸

¸¸

2

Richard M. Murray, Caltech CDSTeam Caltech, Apr 07

Application of existing controls technology in Alice• Receding horizon (optimization-based) control for path

planning with obstacles; ~100 msec iteration rate• Multi-layer sensor fusion: sensor “bus” allows different

combinations of sensors to be used for perceptors + fusion at “map” level

• Low level (inner loop) controls: PID w/ anti-windup (but based on a feasible trajectory from RHC controller)

DGC07 System Architecture

3

FeatureClassificat’n

ElevationMapping

ObstacleDetect/Track

LADAR (6)

Stereo/RoadFinding

GimbaledSensor

World Map

Obstacle Map

Vehicles

PathPlanner

PathFollower

ActuationInterface

TrafficPlanner

MissionPlanner

Vehicle StateEstimator Vehicle

Sensing

Navigation

Systems

ProcessManager

HealthManager

Logging/Visualization Simulation

Properties• Highly modular• Rapidly adaptable• Constantly viable• Robust ???

Linux,TCP/IP, ...

Richard M. Murray, Caltech CDSTeam Caltech, Apr 07

Sensing Bowtie

MapElement serves as “constraint that deconstrains”• Fix the structure of the elements in the world map• Left end: sensors → perceptors → MapElements• Right end: MapElements → environment descriptions → planners

Engineering principle: allow parallel development (people and time) + flexibility• Fixing the map element structure allows 15 people to work simultaneously• We can evolve/adapt our design over time, as we get closer to the race

4

FeatureClassificat’n

ElevationMapping

ObstacleDetect/Track

LADAR (6)

Stereo/RoadFinding

GimbaledSensor

World Map

Obstacle Map

Vehicles

MapElement

Richard M. Murray, Caltech CDSTeam Caltech, Apr 07

Protocol stack based architecture• Planners uses directives/responses to communicate• Each layer is isolated from the ones above and below• Had 4 different path planners under development, two

different traffic planners.

Engineering principle: layered protocols isolate interactions• Define each layer to have a specific purpose; don’t rely

on knowledge of lower level details• Important to pass information back and forth through

the layers; a fairly in an actuator just generate a change in the path (and perhaps the mission)

• Higher layers (not shown) monitor health and can act as “hormones” (affecting multiple subsystems)

Hybrid system control methodology• Finite state automata control interactions between layers

and mode switches (intersection, off road, etc)• Formal methods for analysis of control protocol correctness (post race)

- Eg: make sure that you never have a situation where two layers are in conflict

Planning Hourglass

5

PathPlanner

PathFollower

ActuationInterface

TrafficPlanner

MissionPlanner

Vehicle

Richard M. Murray, Caltech CDSUTC, 27 Oct 2010

Lessons Learned from AliceOnline optimization solves nonlinear control problems• Modern computation allows constrained optimization

problems to be solved online• Solutions exist for situations with more limited computation

(multi-parametric optimization)

Layered control architectures allow more efficient design• Allows for “separation of concerns” between subsystems• Provided a very modular design, capable of rapid (human-

controlled) adaptation

Verification of control protocols is necessary, but hard• Traditional methods of simulation and testing not sufficient• Formal methods not easily applied to “hand designed”

control protocols

New tools for “correct by construction” design are needed• Temporal logic(s) are powerful language for specifying

desired behavior (combined with traditional measures)• New tools are becoming available, but not yet ready for

prime time

6

PathPlanner

PathFollower

ActuationInterface

TrafficPlanner

MissionPlanner

Vehicle

Richard M. Murray, Caltech CDSEECI, Mar 2011

Lecture Schedule

7

Mon Tue Wed Thu Fri

9:00L1: Intro to

Protocol-Based Control Systems

L5: Logic Synthesis

L7: Control-Theoretic Tools

L13: Extensions, Applications and Open Problems

11:00 L2: Automata Theory Computer Lab

L8: Analysis of Control

Protocols

L14: Future Research Directions

12:30 Lunch Lunch Lunch Lunch Lunch

14:00 L3: Linear Temporal Logic

L9: Synthesis of Control

Protocols

L11: TuLiP (RHTLP toolbox)

16:00 L4: Model Checking

L10: Receding Horizon

Temporal Logic Planning

Computer Lab

http://www.cds.caltech.edu/~utopcu/eeci2011.html

Richard M. Murray, Caltech CDSEECI, Mar 2011

Formal Methods for System Verification & SynthesisSpecification using LTL• Linear temporal logic (LTL)

is a math’l language for describing linear-time prop’s

• Provides a particularly usefulset of operators for construc-ting LT properties without specifying sets

Methods for verifying an LTL specification• Theorem proving: use formal

logical manipulations• Model checking: search for paths

that satisfy the system dynamics(transition system) and violate the system specification (LTL formula)- If none, system is correct. Otherwise, return a counter example

Methods for synthesis: paths + finite state automata• Feasible paths: use model checking to find a “counter-example” (= feasible solution)• Finite state automata: determine how to react to environment to satisfy a spec

8

Richard M. Murray, Caltech CDSEECI, Mar 2011

GCDrive FSM Verification (and Synthesis?)

Verification using temporal logic• Model follower, Actuation Interface, DARPA, accModule, transModule in TLC

• Shared variables: state, estop, acc, acc_command, trans, trans_command

Verify the following properties

• ¨((estop = DISABLE) ⇒ ◊¨(state = DISABLED ∧ acc = -1))

• ¨((estop = PAUSE) ⇒ ◊(state = PAUSED ∨ estop = DISABLE))

• ¨((estop = RUN) ⇒ ◊(state = RUNNING ∨ state = RESUMING))

• ¨((state = RESUMING) ⇒ ◊(state = RUNNING ∨ estop = DISABLE ∨ estop = PAUSE))

• ¨((state ∈ {DISABLE, PAUSED, RESUMING, SHIFTING} ⇒ acc = -1)

Richard M. Murray, Caltech CDSEECI, Mar 2011

Receding Horizon Control for Linear Temporal LogicFind planner (logic + path) to solve general control problem

• Can find automaton to satisfy this formula in O((nm|Σ|3) time (!)

Basic idea• Discretize state space into regions { } + interconnection graph• Organize regions into a partially ordered set { }; ⇒ if state starts in , must transition through on way to goal

• Find a finite state automaton satisfying

- Φ describes receding horizon invariants (eg, no collisions)- Automaton states describe sequence of regions we transition

through; is intermediate (fixed horizon) goal- Planner generates trajectory for each discrete transition- Partial order condition guarantees that we move closer to goal

Properties• Provably correct behavior according to spec

10

(ϕinit ∧ �ϕe) =⇒ (�ϕs ∧ ♦ϕg)• φinit = init conditions

• φe = envt description• φs = safety property

• φg = planning goal

Ψi =((v ∈Wi) ∧ Φ ∧ �ϕe) =⇒ (�ϕs ∧ ♦(v ∈Wgi) ∧ �Φ)

W1

W2

W3

W4Wi

Vi

Ai

Wj �ϕg Wi

Wi Wj

Wgi �φg Wi

Wongpiromsarn, Topcu and MIEEE TAC, 2010 (s)

Richard M. Murray, Caltech CDSEECI, Mar 2011 11

Applications

Richard M. Murray, Caltech CDSEECI, Mar 2011 12

Control in an Information Rich World (2001-03)1. Executive Summary

2. Overview of the Field• What is Control?• Control System Examples• Increasing Role of Information-

Based Systems• Opportunities and Challenges

3. Applications, Opportunities & Challenges• Aerospace and Transportation• Information and Networks• Robotics and Intelligent Machines• Biology and Medicine• Materials and Processing• Other Applications

4. Education and Outreach

5. Recommendations

SIAM, 2003

Richard M. Murray, Caltech CDSEECI, Mar 2011 13

Transportation and AerospaceThemes• Autonomy• Real-time, global, dynamic networks• Ultra-reliable embedded systems• Multi-disciplinary teams• Modeling for control

- more than just- analyzable accurate hybrid models

Technology Areas�Air traffic control, vehicle management�Mission/multi-vehicle management�Command & control, human in the loop�Ground traffic control (air & ground)�Automotive vehicle & engine control�Space vehicle clusters�Autonomous control for deep space

Richard M. Murray, Caltech CDSEECI, Mar 2011 14

Information and NetworksPervasive, ubiquitous, convergent networking• Heterogeneous networks merging communica-

tions, computing, transportation, finance, utilities, manufacturing, health, entertainment, ...

• Robustness/reliability are dominant challenges• Need “unified field theory” of communications,

computing, and control

Many applications• Congestion control on the internet• Power and transportation systems• Financial and economic systems• Quantum networks and computation• Biological regulatory networks and evolution• Ecosystems and global change

Control of the network

Control over the network

Richard M. Murray, Caltech CDSEECI, Mar 2011 15

Robotics and Intelligent MachinesWiener, 1948: Cybernetics• Goal: implement systems capable of

exhibiting highly flexible or ``intelligent'' responses to changing circumstances

DARPA, 2003: Grand Challenge• LA to Las Vegas (400 km) in 10 hours or less• Goal: implement systems capable of

exhibiting highly flexible or ``intelligent'' responses to changing circumstances

Richard M. Murray, Caltech CDSEECI, Mar 2011 16

Biology and Medicine“Systems Biology”• Many molecular mechanisms for biological organisms are

characterized• Missing piece: understanding of how network

interconnection creates robust behavior from uncertain components in an uncertain environment

• Transition from organisms as genes, to organisms as networks of integrated chemical, electrical, fluid, and structural elements

Key features of biological systems• Integrated control, communications, computing• Reconfigurable, distributed control, at molecular level

Design and analysis of biological systems• Apply engineering principles to biological systems• Systems level analysis is required• Processing and flow of information is key

Richard M. Murray, Caltech CDSEECI, Mar 2011 17

Materials and ProcessingMulti-scale, multi-disciplinary modeling and simulation• Coupling between macro-scale actuation and micro-scale

physics• Models suitable for control analysis and design

Increased use of in situ measurements• Many new sensors available that generate real-time data

about microstructural properties

Richard M. Murray, Caltech CDSEECI, Mar 2011 18

CDS Panel Recommendations1. Substantially increase research aimed at the integration of

control, computer science, communications, and networking.

2. Substantially increase research in control at higher levels of decision making, moving toward enterprise level systems.

3. Explore high-risk, long-range applications of control to areas such as nanotechnology, quantum mechanics, electromagnetics, biology, and environmental science.

4. Maintain support for theory and interaction with mathematics, broadly interpreted.

5. Invest in new approaches to education and outreach for the dissemination of control concepts and tools to non-traditional audiences.

Richard M. Murray, Caltech CDSEECI, Mar 2011 19

Top 10 Challenge Problems in Control (2005)1. World Cup Robotic Soccer Team. Design a robotic soccer team that is good enough that it can compete against humans in the World Cup (and win!).

2. InternetRT. Redesign the Internet so that it could be used to provide real-time (RT) connections between sensors, actuators, and computation that had arbitrary geographic locations.

3. Dynamically Reconfigurable Air Traffic Control. Design the air traffic control system so that passengers always got to their destination on time, with a plane that was always 90+% full, and with no delays due to weather anyplace in the country except your departure or arrival city.

4. Human Life Stabilization Bay. Design a medical system that, when connected to a human through biometric sensors and drug delivery systems, can diagnose a human being and keep it alive indefinitely.

5. Redesign the Feedback Control System of a Bacteria. Can we redesign the control systems in bacteria (including implementation!) so that we can program their behaviors in response to external stimuli?

6. Control 101. Develop a single, unified curriculum for teaching undergraduate control that is a required course for all engineering disciplines (including computer science!).

7. Packet-Based Control Theory. Develop a theory for control in which the basic input/output singles are data packets that may arrive at variable times, not necessarily in order, and sometimes not at all.

8. Slow Computing. Build a computer capable of powering a PDA out of devices with 10 msec (100 Hz) switching times.

9. Write a 100,000+ line program that works correctly on first execution.

Richard M. Murray, Caltech CDSUTC, 27 Oct 2010

Some Important Trends in Control in the Last Decade(Online) Optimization-based control• Increased use of online optimization (MPC/RHC)• Use knowledge of (current) constraints &

environment to allow performance and adaptability

Layering and architectures• Command & control at multiple levels of abstraction• Modularity in product families via layers

Formal methods for analysis, design and synthesis• Combinations of continuous and discrete systems• Formal methods from computer science, adapted for

cyberphysical systems

Components → Systems → Enterprise• Movement of control techniques from “inner loop” to

“outer loop” to entire enterprise (eg, supply chains)• Use of systematic modeling, analysis and synthesis

techniques at all levels• Integration of “software” with “controls”

20

Richard M. Murray, Caltech CDSLIDS, Nov 09

Control of Complex Systems

Alice• 30 P4 cores @ 2+ GHz• 30G mem, 3 Gb/s net• 8 LADAR, 10 cameras• 200k+ lines of code• 50 engineers, 1.5 years• 600 miles of self-driving• Dominant challenges:

(design for) verification and robustness

Drosophila• 300k neurons @100 Hz• ~600-1000 sensors• Most neurons are

dedicated to sensor processing

• Architecture (?)

• Dominant challenges: decoding organization and architecture

Neutrophil• DNA: 3G base pairs• 20k genes, ~3kb each • Transcription: 75 bp/

sec => 40 sec “clock”• Life span: 12 hours• Architecture (?)

• Dominant challenges: (lack of) modularity, stochastic program’g

21

Sawyer Fuller (Caltech), Nov 08

Gwenyth Card (Caltech), Apr 06

Richard M. Murray, Caltech CDSEECI, Mar 2011

Agent dynamics - continuous

Agent mode (or “role”) - discrete• encodes internal state +

relationship to current task

• Transition

Communications graph• Encodes the system information flow

• Neighbor set

Communications channel• Communicated information can be lost,

delayed, reordered; rate constraints

• γ = binary random process (packet loss)

Task• Encode task as finite horizon optimal

control + temporal logic (assume coupled)

Strategy• Control action for individual agents

Decentralized strategy

• Similar structure for role update

Systems to Enterprise: Cooperative Control

22

N i(x,α)

J =� T

0L(x,α, u) dt + V (x(T ),α(T )),

ui = γ(x,α) {gij(x,α) : ri

j(x,α)}

αi � =

�rij(x,α) g(x,α) = true

unchanged otherwise.

α ∈ A

α� = r(x,α)

G

MJGCD, 2007

xi = f i(xi, ui) xi ∈ Rn, ui ∈ Rm

yi = hi(xi) yi ∈ Rq

yij [k] = γyi(tk − τj) tk+1 − tk > Tr

ui(x,α) = ui(xi,αi, y−i,α−i)

y−i = {yj1 , . . . , yjmi}jk ∈ N i mi = |N i|

(ϕinit ∧ �ϕe) =⇒ (�ϕs ∧ ♦ϕg)

Richard M. Murray, Caltech CDSEECI, Mar 2011

Straw and Dickson (Caltech)

23

Networked Feedback Systems in Biology - Insects

Different architecture than engineering• Large collection of diverse sensors

(many more than required)• Very slow computation with lots of

parallel pathways

Card and Dickinson (Caltech) Tensor product

y

+!

g

y

f , !

fk = Buk

vw

vyu

! k = Auk

vw

vyu

w

Sean Humbert (Caltech/U. Maryland)

S. Fuller (Caltech)

Censi, Han, Fuller, Straw and MICRA09, CDC09

Stabilization of a goal image/optical flow• Compute forces and torques based on

spatial integration of image space error• Bioplausible: implement w/ slow computing• Bootstrapable: can learn operator weights

by estimating velocity given image + torques

Richard M. Murray, Caltech CDSEECI, Mar 2011

http://www.genomics.princeton.edu/ryulab

24

Networked Feedback Systems in Biology - Cells

A. Prindle (Caltech/UCSD)

What’s different about biological systems• Complexity - biological systems are much

more complicated than engineered systems

• Communications - signal representations are very different (spikes, proteins, etc)

• Uncertainty - very large uncertainty in components; don’t match current tools

• Evolvability - mutation, selection, etc

Potential application areas for control tools• System ID - what are the appropriate

component abstractions and models?

• Analysis - what are key biological feedback mechanisms that lead to robust behavior?

• Design - how to we (re-)design biological systems to provided desired function?

• Fundamental limits - what are the limits of performance and robustness for a given biological network topology?

Richard M. Murray, Caltech CDSBoeing, 29 Sep 2010

Control of Complex Systems

Alice• 30 P4 cores @ 2+ GHz• 30G mem, 3 Gb/s net• 8 LADAR, 10 cameras• 200k+ lines of code• 50 engineers, 1.5 years• 600 miles of self-driving• Dominant challenges:

(design for) verification and robustness

Drosophila• 300k neurons @100 Hz• ~600-1000 sensors• Most neurons are

dedicated to sensor processing

• Architecture (?)

• Dominant challenges: decoding organization and architecture

Neutrophil• DNA: 3G base pairs• 20k genes, ~3kb each • Transcription: 75 bp/

sec => 40 sec “clock”• Life span: 12 hours• Architecture (?)

• Dominant challenges: (lack of) modularity, stochastic program’g

25

Sawyer Fuller (Caltech), Nov 08