CMPUT 412 Introduction. Csaba Szepesvári University of Alberta

Transcription

1 CMPUT 412 Introduction Csaba Szepesvári University of Alberta

2 Table of contents Admin Robots Basics of control Robot design

3 Admin Teams: Not yet assembled (next week) Rotational scheme First lab: Learn about NXT Evaluation update: Team members evaluate each others performance Fill out a form (who did what) Team of N can get total of N*5 points (5: good), consensus 10% - class participation (M=0.9*T + 0.1*cp)

4 Robots Manipulators Mobile robots

5 Manipulators

6 Here: Mobile Robots Mobile: Changes its point of reference with respect to the environment Consequences: Advantages Problems: Where am I? Hidden state, aliasing,.. What is happening? Unstructured environments..

7 Mobile robots: Advantages Provide specialized access hazard, environment (no air, etc.), distance/time (Mars) Reduce operating costs lower overheads, reduced maintenance costs (gentler treatment of the machinery) Increase productivity permanent availability - more hours; higher throughput Improve product quality accuracy, consistency New human services! human interactivity

8 Applications Medical Cleaning Commercial Household Sales (automated gas pump in Japan) Agriculture (Demeter) Forestry (pruning Xmas trees) Lawn care (golf courses) Hazardous (highpower line inspection, pipe inspection, building inspection) Mining/Excavation Space / Undersea Military Security Personal (handicapped home) Entertainment (Sony s RoboDog)

9 AGV by Volvo Automatic Guided Vehicle of VOLVO Goal: transport motor blocks from on assembly station to an other Navigation: guided by an electrical wire installed in the floor able to leave the wire to avoid obstacles. Over 4000 AGV only at VOLVO s plants!

10 Helpmate Goal: transportation in hospitals Navigation: Look at the ceiling with a camera (landmarks) + corners

11 BR700 Cleaning robot Goal: Cleaning Navigation: sonar system + gyro Kärcher Inc.

12 ROV Tiburon underwater robot Goal: Underwater archeology Control: Hovering only Navigation: Teleoperated

13 The Pioneer Goal: Explore sarcophagus at Chernobyl Control: Teleoperated

14 Pioneer I. Goal: Research Stanford Robotics Institute

15 The B21 Robot Goal: Research Manufacturer: Real World Interface

16 Forester robot Moving wood out of forest Control: Human, leg coordination automated Pulstech

17 Tube inspection Goal: sewage tube inspection and reparation Control: teleop. HÄCHER Goal: Airduct inspection Control: Teleop. EFPL EFPL robot (4mins)

18 Sojourner Mars robot Goal: Explore Mars Control: Teleoperation (1cm/sec, 10min delay), some obstacle avoidance (2 mins)

19 Nomad Goals: autonomous search of Antarctic meteorites Developer: CMU, NASA 1997

20 Honda

21 Aibo from Sony Stereo mics Color camera 10 cm

22 Vacuum Cleaners Dyson DC06 (cancelled 2004) irobot Roomba Electrolux Trilobite

23 The Cye Personal Robot Goal: Vacuum cleaner, trailer Remote controlled (PC) Two wheels!

24 How to control robots? Controlled systems Controllers Controller goals

25 Abstract control model Environment Sensations (and reward) actions Controller = agent Perception-action loop

26 Zooming in.. external sensations memory reward agent state internal sensations actions

27 Mathematical model Plant (controlled object): x t+1 = f(x t,a t ) + v t x t : state Y t = g(x t ) + w t y t : sens/obs State: Sufficient statistics for the future Independently of what we measure..or.. Relative to measurements (goals) Controller a t = F(y 1,y 2,,y t ) a t : action/control => PERCEPTION-ACTION LOOP CLOSED-LOOP CONTROL Design problem: F =?

28 What: Goals of a robot Goal directed behavior Environmental (manipulation) goals Brew coffee Pour water into the coffee machine Pick up a cup Position goals: Get to a place Deliver packages Homeostasis Reward hypothesis: t=1 T r t max Homeostasis: z t Z, z t : internal signals

29 Competencies: What -> How High level goals environmental (manipulation) goals homeostasis position goals Supporting competencies knowledge of environment internal state position

30 How: Controllers Feedforward: a 1,a 2, is designed ahead in time??? Feedback: Purely reactive systems: a t = F(y t ) Why is this bad? Feedback with memory: m t = M(m t-1,y t,a t-1 ) ~interpreting sensations a t = F(m t ) decision making: deliberative vs. reactive

31 Feedback controllers Plant: x t+1 = f(x t,a t ) + v t y t+1 = g(x t ) + w t Controller: m t = M(m t-1,y t,a t-1 ) a t = F(m t ) m t x t : state estimation, filtering difficulties: noise,unmodelled parts How do we compute a t? With a model (f ): model-based control..assumes (some kind of) state estimation Without a model: model-free control

32 A control scheme for mobile robots Knowledge, Data Base Mission Commands Localization Map Building "Position" Global Map Cognition Path Planning Environment Model Local Map Path Perception Information Extraction Raw data Sensing Path Execution Actuator Commands Acting Motion Control Real World Environment (Siegwart, Nourbakhsh)

33 How to Design Robots? Mechanics - Body (chassis, actuation) Energy - Power Inputs - Sensors Brain - Control

34 Deconstruction.. Mechanical chassis load bearing & distribution, suspension effectors (mobility) wheels/gears, legs, tuna!, helicopter rotors pan/tilt units; reconfiguration units (pipe crawlers, throw-in-bldg) Power Batteries Solar Nuclear Generators (gas)

35 Deconstruction../2 Sensing internal sensors - voltages, load, encoders position estimation sensors measuring relative position of robot compass, odometer, gyro, accelerometer, INS, GPS, startscope, sunscope environmental sensors - measuring characteristics of the external world bumpers, sonar, infrared, laser-rangefinder, radar, CCD s, Mars soil sampler weird ones: torque/force transducers, limit switches Control in most general case: y* a multilevel control versus 1-level control: goal motion; then achieving it... (outdoor mobile robot)

36 Big issue questions Scaling up & scaling down! Sensors: how much is enough Exceptions/errors What are they? What does it mean??? Completeness/optimality What do they mean? Deliberation vs. reactivity Models/lookahead required? Memory?

37 Big issue questions/2 Representation If we want one? transparency/opacity How do we do uncertainty? Model explicitly, engineer around? Better sensors? Better effectors? Learning (how does it fit in?) Off-line learning / parameter tweaking, RL? Robot teams?? Architecture! tiered, parallel, sequential, etc.

38 Summary Mobile robots Advantages and challenges Working with (mobile) robots requires understanding of: Engineering issues Control issues Signal processing issues Closed-loop, FB vs. FF control, memory, state, state-estimation/filtering, modelbased/-free, deliberative vs. reactive