What is a robot? Lectre 2: Basics CS 344R/393R: ics Benjamin Kipers A robot is an intelligent sstem that interacts with the phsical environment throgh sensors and effectors. Toda we discss: Abstraction Sensor errors Color perception sensors effectors Remember the Amigobot? Sonar sensors: front (6), back (2) Camera Passive gripper Differential drive (right/left wheel) Odometr Wireless commnication Describing the Amigobot State vector: x = (x,,") T The tre state is not known to the robot. Sense vector: = (s 1,s 2,s 3,s 4,s 5,s 6,s 7,s 8,o L,o R ) T Sonars and odometr Pls sensor featres from camera Motor vector: = (v L,v R ) T Left-wheel, right-wheel These are fnctions of time: x(t), (t), (t) Derivative notation: x = dx /dt Modeling Interaction sensors = H i () x = F(x,) = G(x) effectors The Uniccle Model For the niccle, = (v, ω) T, where v is linear velocit ω is anglar velocit # x & # v cos" & % ( % ( x = % ( = F (x, ) = % v sin" ( % $ ( % ( "' $ ) ' A sefl abstraction for mobile robots. 1
The Amigobot is (like) a Uniccle # x & # v cos" & % ( % ( x = % ( = F (x, ) = % v sin" ( % $ ( % ( "' $ ) ' Amigobot motor vector: = (v L, v R ) v = (v R + v L )/2 (mm/sec) " = (v R #v L )/ B (rad/sec) where B is the robot wheelbase (mm). Abstracting the Model sensors effectors Abstracting the Model control law ' sensor featre ' motor command Abstracting the Model control law ' sensor featre ' motor command Abstracting the Model '' control law ' sensor featre ' motor command Abstracting the Model B implementing sensor featres and control laws, we define a new robot model. New sensor featres New motor signals The robot s environment changes from continos, local, ncertain to reliable discrete graph of actions. (For example. Perhaps. If o are lck.) We abstract the Aibo to the Uniccle model Abstracting awa joint positions and trajectories 2
A Topological Abstraction For example, the abstracted motor signal cold select a control law from: TrnRight, TrnLeft, Rwall, Lwall, Midline The abstracted sensor signal cold be a Boolean vector describing nearb obstacles: [L, FL, F, FR, R] The continos environment is abstracted to a discrete graph. Discrete actions are implemented as continos control laws. Mobile robots Tpes of s Or class focses on these. Atonomos agent in nfamiliar environment. maniplators Often sed in factor atomation. Programmed for perfectl known workspace. al monitoring robots Distribted sensor sstems ( motes ) And man others Web bots, etc. Tpes of Sensors Range-finders: sonar, laser, IR Odometr: shaft encoders, ded reckoning Bmp: contact, threshold Orientation: compass, accelerometers GPS Vision: high-res image, blobs to track, motion Sensor Errors: Accrac and Precision accrate precise both Related to random vs sstematic errors Sonar vs Ra-Tracing Sonar doesn't perceive distance directl. It measres "time to echo" and estimates distance. Sonar Sweeps a Wide Cone. One retrn tells s abot man cells. Obstacle cold be anwhere on the arc at distance D. The space closer than D is likel to be free. Two Gassians in polar coordinates. 3
Sonar chirp fills a wide cone Data on sonar responses Sensing a flat board (Left) or pole (Right) at different distances and angles. For the board (2'x8'), secondar and tertiar lobes of the sonar signal are important. Speclar Reflections in Sonar Mlti-path (speclar) reflections give spriosl long range measrements. Exploring a Hallwa with Sonar A Usefl Heristic for Sonar Short sonar retrns are reliable. The are likel to be perpendiclar reflections. Lassie sees the world with a Laser Rangefinder 180 ranges over 180 planar field of view Abot 13 above the grond plane 10-12 scans per second 4
Laser Rangefinder Image 180 narrow beams at 1º intervals. Ded ("Dead") Reckoning From shaft encoders, dedce (Δx i, Δ i, Δθ i ) Dedce total displacement from start: (x,,") = (0,0,0) + $ (#x i,# i,#" i ) How reliable is this? It s prett bad. Each (Δx i, Δ i, Δθ i ) is OK. i Cmlative errors in θ make x and nreliable, too. Odometr-Onl Tracking: 6 times arond a 2m x 3m area Hman Color Perception Perceived color is a fnction of the relative activation of three tpes of cones in the retina This will be worse for the Aibo walking. The Gamt of the Hman Ee Gamt: the set of expressible colors RGB: An Additive Color Model Three primar colors stimlate the three tpes of cones, to achieve the desired color perception. 5
Color Perception and Displa Onl some hman-perceptible colors can be displaed sing three primaries. HSV: He-Satration-Vale HSV attempts to model hman perception L * a * b * (CIELAB) is more perceptall accrate Lightness; a * : red-green axis; b * : ellow-ble Aibo Uses the YUV Color Model RGB rotated Y: Lminance U-V: he Used in PAL video format To track, define a region in color space. See Tekkots ttorial Or Goals for ics From nois low-level sensors and effectors, we want to define reliable higher-level sensor featres, reliable control laws for meaningfl actions, reliable higher-level motor commands. Understand the sensors and effectors Especiall inclding their errors Use abstraction 6