Survey
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project
Download fast - Control and Dynamical Systems
Document related concepts
no text concepts found
Transcript
Mechanics+ Gravity + Light + vision Balancing an inverted pendulum Act p ln T j 2 d 2 0 p 1 z p p ln z p delay + Neuroscience vision delay Balancing an inverted pendulum Act 2p ln T j 2 d p 2 0 p 1 Law #1 : Mechanics (instead of chemistry) Law #2 : Gravity (instead of autocatalysis) 2 M m x ml cos sin u easy hard x cos l g sin 0 y x l sin M m x ml u x l g 0 linearize y x l 2 M m x ml cos sin u x cos l g sin 0 y x l sin linearize M m x ml u x l g 0 y x l m l M Law #3 : Light M m x ml u x l g 0 y x l Why? vision delay hard harder Act Easy to prove using simple models. noise error eye vision slow l E T j N delay Control 1 p l Act Law #4 : 2p ln T j 2 2 0 p 1 d p 8 1 p l Fragility 4 p .3s 2 1 .05 .1 .2 .5 Length l (meters) 1 Law #4 : 2p ln T j 2 2 0 p 1 d p 8 1 p l Fragility 4 p .3s 2 .1s 1 .05 .1 .2 .5 Length l (meters) 1 Crashes can be made rare with active control. What is sensed matters. hard harder hardest! Why? Easy to prove using simple models. What is sensed matters. hard Unstable poles harder hardest! Unstable zeros 2p ln T j 2 2 0 p 1 z p d p ln z p 8 Fragility p l0 l 1 4 l0 l z p ln 2 z p .1s .3s 1 .05 .1 .2 .5 Length (meters) 1 1.2 easy 3.5 1 3 Measure Length l0, m 0.8 2.5 0.6 hard 2 0.4 hard 1.5 1 0.2 fragile (hard) 0.4 0.6 0.8 1 Length l to CoM, m robust (easy) Completing the story vision Balancing an inverted pendulum Act Mechanics+ Gravity + Light + p ln T j 2 d 2 0 p 1 z p p ln z p delay + Neuroscience Explain this amazing system. Slow Vision Fast Flexible Inflexible • Neuroscience motivation Robust vision with motion Motion Vision Experiment • • Motion/vision control without blurring Which is easier? Why? • Mechanism • Tradeoff Slow vision Fast VOR Slow Mechanism Vestibular Ocular Reflex (VOR) vision Tradeoff VOR Fast Flexible Inflexible eye Act Slow vision slow delay vision Fast Flexible Inflexible eye vision fast slow Act Slow delay vision VOR Fast Flexible Inflexible eye fast Act Vestibular Ocular Reflex (VOR) Slow VOR Fast Flexible Inflexible eye fast Act Slow VOR Fast Flexible Inflexible eye vision fast slow Act Slow delay vision VOR Fast Flexible Inflexible eye vision slow Act Highly evolved (hidden) architecture Slow vision Fast Illusion VOR Flexible Inflexible delay eye vision fast slow Act delay Slow vision Fast Illusion VOR Flexible Inflexible Fragile Slow Fast Ideal Flexible Inflexible Fragile Slow Fast Architecture (constraints that deconstrain) Ideal Flexible Inflexible eye vision fast slow Act Slow delay vision VOR Fast Flexible Inflexible VOR eye vision slow Act Balancing an inverted pendulum Act delay fast Control noise error eye vision slow l E T j N delay Control 1 p l Act Law #4 : 2p ln T j 2 2 0 p 1 d p 8 1 p l Fragility 4 p .3s 2 1 .05 .1 .2 .5 Length l (meters) 1 What is sensed matters. hard harder hardest! Why? Easy to prove using simple models. 8 Fragility p l0 l 1 4 l0 l z p 2 ln z p .1s 1 .05 .1 .2 .5 Length (meters) 1 1.2 easy 3.5 1 3 Measure Length l0, m 0.8 2.5 0.6 hard 2 0.4 hard 1.5 1 0.2 fragile (hard) 0.4 0.6 0.8 1 Length l to CoM, m robust (easy) Slowest 3D + motion eye vision fast Act Slow color vision delay vision VOR Fast Flexible slow Inflexible B&W (luminence only): 3D, motion, and action 3D + motion eye vision fast Act Slow delay vision VOR Fast Flexible slow Inflexible Stare at thegood. intersection This is pretty Stare at thegood. intersection This is pretty Slowest 3D + motion eye vision fast Act Slow color vision delay vision VOR Fast Flexible slow Inflexible Seeing is dreaming color vision Slow vision VOR Fast Flexible Inflexible slower Objects Slow 3d+motion B&W slow Fast Mixed Flexible fast VOR Inflexible Not sure how to draw this… slower Objects Slow Faces 3d+motion B&W slow Fast Mixed Flexible fast VOR Inflexible eye Seeing is dreaming slow Act Highly evolved (hidden) architecture Slow vision Fast Illusion VOR Flexible Inflexible vision delay Conscious perception Seeing is dreaming 3D +time Simulation + complex models (“priors”) Conscious perception Prediction &Control errors Other dimensions? Slow Cheap Fast Costly Flexible Inflexible General Special Requirements on systems and architectures accessible accountable accurate adaptable administrable affordable auditable autonomy available credible process capable compatible composable configurable correctness customizable debugable degradable determinable demonstrable dependable deployable discoverable distributable durable effective efficient evolvable extensible fail transparent fast fault-tolerant fidelity flexible inspectable installable Integrity interchangeable interoperable learnable maintainable manageable mobile modifiable modular nomadic operable orthogonality portable precision predictable producible provable recoverable relevant reliable repeatable reproducible resilient responsive reusable robust safety scalable seamless self-sustainable serviceable supportable securable simplicity stable standards compliant survivable sustainable tailorable testable timely traceable ubiquitous understandable upgradable usable Sustainable robust + efficient accessible accountable accurate adaptable administrable affordable auditable autonomy available compatible composable configurable correctness customizable debugable degradable determinable demonstrable dependable deployable discoverable distributable durable effective manageable safety mobile scalable modifiable seamless modular self-sustainable nomadic serviceable operable supportable efficient orthogonality securable evolvable portable simple extensible precision stable fail transparent predictable standards fast producible survivable fault-tolerant provable sustainable fidelity recoverable tailorable flexible relevant testable inspectable reliable timely installable repeatable traceable Integrity reproducible ubiquitous interchangeable resilient understandable interoperable responsive upgradable learnable reusable usable maintainable robust PCA Principal Concept Analysis accessible dependable manageable safety accountable deployable mobile scalable accurate discoverable modifiable seamless adaptable distributable modular self-sustainable administrable durable nomadic serviceable affordable effective operable supportable auditablefragileefficient orthogonalit securable Simple autonomy evolvable y simple available extensible dichotomous portable stable compatible fail precision standards tradeoff pairs composable transparent predictable survivable configurable fast producible sustainable correctness fault-tolerant provable tailorable robustfidelity customizable recoverable testable debugable flexible relevant timely degradable inspectable reliablewasteful traceable efficient determinable installable repeatable ubiquitous demonstrable Integrity reproducible understandable interchangeabl resilient upgradable e responsive usable interoperable reusable robust learnable maintainable UG biochem, math, control theory Chandra, Buzi, and Doyle Most important paper so far. Hard tradeoff in glycolysis is • robustness vs efficiency • absent without autocatalysis • too fragile with simple control • plausibly robust with complex control fragile 1 10 too fragile z p z p z ln S j 2 d 2 0 z 1 complex 0 z p ln z p No tradeoff 10 -1 10 k 0 10 1 10 expensive VOR 8 eye vision slow Act Fragility 4 p delay fast Balancing an inverted pendulum Act l0 l 1 Control ln l0 l z p2 z p .1s 1 .05 .1 1 .2 .5 Length (meters) 1 10 fragile too fragile z p z p complex No tradeoff 0 10 -1 10 k 0 10 1 10 expensive VOR 8 eye Fragility vision slow Act 4 p delay fast Balancing an inverted pendulum l0 l 1 Control ln l0 l z p2 z p .1s 1 Act .05 .1 1 .2 .5 Length (meters) 1 10 fragile Slow too fragile z p z p complex Fast No tradeoff 0 Flexible General Inflexible Special 10 -1 10 k 0 10 1 10 expensive VOR 8 eye vision slow Act p delay fast Balancing an inverted pendulum Fragility ln Control z p z p l0 l 1 4 l0 l 2 .1s 1 Act .05 Slow Fast Flexible General Inflexible Special .1 .2 .5 Length (meters) 1 Universal laws and architectures (Turing) Slow Architecture (constraints that deconstrain) Fast Flexible Inflexible General Special Computational complexity Really slow Slow Fast Undecidable Flexible/ General Decidable NP Inflexible/ Specific P Sustainable robust + efficient accessible accountable accurate adaptable administrable affordable auditable autonomy available compatible composable configurable correctness customizable debugable degradable determinable demonstrable dependable deployable discoverable Issues distributable durableFast effective manageable safety mobile scalable modifiable seamless modular self-sustainable nomadic serviceable operable supportable Robust orthogonality securable efficient evolvable portable simple Flexible extensible precision stable fail transparent standards Efficient predictable fast producible survivable Stochasticsprovable sustainable fault-tolerant fidelity recoverable tailorable Memory flexible relevant testable inspectable reliable timely installable repeatable traceable Integrity reproducible ubiquitous interchangeable resilient understandable interoperable responsive upgradable learnable reusable usable maintainable robust Weaknesses so far • Some flaws as presented • See if we can find the flaws and fix them • What could be improved? – Model – Theory – Experiment • Suggestions? Model? • 1 dimension with 4 states? • What about the other 2 dimensions? • Let’s imagine (but not derive) a 10 state model and see what would happen • New issues arise noise error eye x3 E T j N x1 Act x2 vision delay Control noise error eye x3 easy x1 E T j N hard Act x2 easy vision delay Control noise error eye x3 easy x1 E T j N hard Act x2 easy vision delay Control Model to Theory? • • • • Linearization? Mostly ok. Actuation and sensing, mostly ok. Noise model? Needs work. Why? Noise and delay is from measuring distance using stereopsis . How to model this? • What about all the detailed physiology of muscle, joints, bone, nerves, etc? • Need layered control architectures. Theory • Analytic results are not scalable • Aim not analytic formulas but tractable algos Main lessons • Theory: hard limits on closed loop performance, aggravated by – Instability (unstable poles) – Delay – Unstable zeros • Neuroscience specific Instabilities in technology • Efficiency • Autocatalysis Select instabilities in biology Working backwards • Society/agriculture/weapons/etc • Bipedalism • Maternal care • Warm blood • Flight • Mitochondria • Translation (ribosomes) • Glycolysis (see 2011 Science paper)
Related documents