Introduction to Embedded and Real-Time Systems W10: Hardware Design Choices and Basic Control Architectures for Mobile Robots

Transcription

1 Introduction to Embedded and Real-Time Systems W10: Hardware Design Choices and Basic Control Architectures for Mobile Robots

2 Outline Hardware design choices Hardware resource management Introduction to control architectures for mobility Examples of different reactive control architectures The obstacle avoidance example

3 Hardware Design Choices

4 Rationale Depending on the application, the appropriate components must be chosen. On of the key design choices is that of the microcontroller (focus of the next slides) Energy is one of the key bottleneck for autonomous operation of embedded systems and component choice as well as poweraware algorithms must be implemented for fulfilling a given mission in an energy-efficient way

5 Application Examples The following applications have fundamentally different requirements: Wireless sensor networks (e.g., environmental monitoring) Acceleration monitoring (e.g., detection of laptop failing) Sound source localization (e.g., robotic attention) Vision-based monitoring (e.g., surveillance)

6 e-puck Capabilities

7 Wireless Sensor Networks Ex. environmental monitoring, see sensorscope.epfl.ch Requirements: - Very low sampling frequency: sub Hz to a few tens of Hz - Ultra low power consumption: 25mW - Energy: solar panel combined with rechargeable battery - Low bandwidth radio link

8 Acceleration monitoring Example application: detection of laptop failing, see Requirements: - Low sampling frequency (1 khz) - Typical A/D conversion - Low power consumption not crucial - Embedded in a device (e.g. laptop)

9 Sound Source Localization Example application: acoustic attention in robots, see -> robot -> sound sensor and lecture notes week 7 Requirements: - Localization based on times of arrival - Medium sampling frequency (30-50 khz) - Typical A/D conversion - Medium to low power consumption - Sharing of on-board robot resources typically used for other parallel functionalities

10 Surveillance using Computer Vision Requirements: - Very large data flow (tens of Mbit/s) - Corresponding signal processing needed - Corresponding memory needs for image storing - Power consumption depends on application and deployment peculiarities

11 Some Hints and Techniques for Hardware Resource Management

12 Wireless Sensor Networks Technical solutions: - Sampling of the continous-time analog sensors using the integrated A/D converter - Typically parallel, low-frequency sampling for environmental variables

13 Wireless Sensor Networks Technical solutions: - Exploit typical assembly functions to place the microcontroller in an idle status and wake up when needed - Dedicated hardware design can be done to do the same with further modules/chips of the system

14 Source: ISI & DARPA PAC/C Program Sensor Node Energy Roadmap Average Power (mw) 10,000 1, Deployed (5W) PAC/C Baseline (.5W) (50 mw) (1mW) Rehosting to Low Power COTS (10x) -System-On-Chip -Adv Power Management Algorithms (50x)

15 Source: ISI & DARPA PAC/C Program Communication/Computation Technology Projection Communication Computation 1999 (Bluetooth Technology) (150nJ/bit) 1.5mW* 2004 (5nJ/bit) 50uW ~ 190 MOPS (5pJ/OP) Assume: 10kbit/sec. Radio, 10 m range. Large cost of communications relative to computation continues

16 Acceleration monitoring Requirements: - Typical A/D conversion: sampling of the continous time analog accelerometer (3 axes) using the integrated A/D converter - Low power consumption not crucial - Embedded in a device (e.g. laptop)

17 Acceleration monitoring - Sampling frequency typically a function of the application and of the accelerometer characteristics

18 Sound Source Localization Requirements: - Medium sampling frequency - E.g.: robot dimension 7.5 cm microphone max inter-distance 5.5 cm speed of sound in air 340 m/s travel time micro-to-micro 0 (orthogonal) to 160 µs (aligned) 6 khz min to max possible on the device - 2 micros: 100 KHz max e.g. 44 khz 23 µs 8 mm resolution but possible aliasing on a plane (dual localization) - 3 micros: 100 khz max e.g. 28 khz 36 µs 12 mm but no aliasing on a plane (unique localization)

19 Sound Source Localization Requirements: - Sharing of on-board robot resources typically used for other parallel functionalities low battery detection poweron led low battery indicator extension connectors 3.3V regul. to all devices 2x stepper motors programming/ debug connector body light RS232 connector dspic 30F MHz 8x IR prox. ON - OFF IR remote control bluetooth radio link RESET running mode selector 10Mhz clock 2.5V 1.8V regul. 8x Red Leds CMOS color camera 3D accelerometer battery protection LiION Battery 3.6V 1.4Ah audio codec 3x micro speaker

20 Surveillance using Computer Vision Requirements: - Very large data flow (tens of Mbit/s) Processing: - Pixels H x V x RGB x fps x 480 x 3 x 30 = 27Mbytes/second - The dspic can execute max 15MIPS (millions of instructions/second) Memory - One image RBG (8,8,8 bits) of 640x480 use 922kbytes - Our dspic has 8kbytes of RAM (Random Access Memory), for variables - Full image acquisition impossible e-puck microcontroller

21 Surveillance using Computer Vision - Possible workaround on e-puck (see also lecture notes week 8): downsampling - 8 fps grayscale, 4 fps color - Image of 1800 pixels (42x42, 80x20)

22 Control Architectures for Mobile Robots

23 Perception-to-Action Loop sensors Reactive (e.g., nonlinear transform, single loop) Reactive + memory (e.g. filter, state variable, multi-loops) Deliberative (e.g. planning, multi-loops) actuators Perception Computation Action Environment

24 Reactive Architectures: Proximal vs. Distal in Theory Proximal: close to sensor and actuators very simple linear/nonlinear operators on crude data high flexibility in shaping the behavior Difficult to engineer in a human-guided way; machine-learning usually perform better

25 Reactive Architectures: Proximal vs. Distal in Theory Distal architectures Farer from sensor and actuators More elaborated data processing (e.g., filtering) Less flexibility in shaping the behavior Easier to engineer in a human-guided way the basic block (handcoding); more difficult to compose the blocks in the right way (e.g., sequence, parallel, )

26 Reactive Architectures: Proximal vs. Distal in Practice A whole blend! Four classical examples of reactive control architecture for solving the same problem: obstacle avoidance. Two proximal: Braitenberg and Artificial Neural Network Two distal: Subsumption and Motor Schema, both behavior-based

27 Ex. 1: Braitenberg s Vehicles light sensors symmetry axis motors a 2b 3a 3b Work on the difference (gradient) between sensors Originally omni-directional sensors but work even better with directional sensors + excitation, - inibition; linear controller (out = signed coefficient * in) Symmetry axis along main axis of the vehicle (----) Originally: light sensors; works perfectly also with proximity sensors (3c?) See also Lab+hwk 3

28 Ex. 2: Artificial Neural Network N i f(x i ) O i output neuron N with sigmoid transfer function f(x) O = w i ij 2 f ( x) = 1 x I j 1+ e synaptic m weight input x i = f j= 1 ( x i w ij I ) j + I 0 S 1 S 2 S 3 S 4 S 5 S 8 S 7 S 6 M 1 M 2 inhibitory conn. excitatory conn.

29 Ex. 3: Rule-Based Rule 1: if (proximity sensors on the left active) then turn right Rule 2: if (proximity sensors on the right active) then turn left Rule 3: if (no proximity sensors active) then move forwards

30 Subsumption Architecture Rodney Brooks 1986, MIT Precursors: Braitenberg (1984), Walter (1953) Behavioral modules (basic behaviors) represented by Augmented Finite State machines (AFSM) Response encoding: predominantly discrete (rule based) Behavioral coordination method: competitive (priority-based arbitration via inhibition and suppression)

31 Subsumption Architecture Sense Model Plan Act Modify the World Create Maps Discover Avoid Collisions Move Around Classical paradigm (serial); emphasis on deliberative control Subsumption (parallel); emphasis on reactive control

32 Subsumption Architecture: AFSM Reset Suppressor Input lines I R Behavioral Module S Output lines Inhibitor Inhibitor: block the transmission Suppressor: block the transmission and replace the signal with the suppressing message

33 Ex. 4: Behavior-Based with Subsumption sensors Obstacle avoidance Wander 1 actuators 2 S (1 suppresses and replaces 2)

34 Evaluation of Subsumption + Support for parallelism: each behavioral layer can run independently and asynchronously (including different loop time) + HW retargetability: can compile down directly to programmable-array logic circuitry - Hardwiring mean less run time flexibility - Coordination mechanisms restrictive ( black or white ) - Limited support for modularity (upper layers design cannot be independent from lower layers).

35 Motor Schemas Ronald Arkin 1987, Georgia Tech Precursors: Arbib (1981), Khatib (1985) Parametrized behavioral libraries (schemas) Response encoding: continuous using potential field analog Behavioral coordination method: cooperative via vector summation and normalization

36 Motor Schemas sensors S 1 PS 1 MS 1 PS 2 S 2 vector Σ actuators S 3 PSS 2 PSS 1 PS 3 MS 2 PS: Perceptual Schema PSS: Perceptual Subschema MS: Motor Schema S: sensor

37 Ex. 5: Behavior-Based with Motor Schemas Detect-obstacles Avoid-obstacle sensors Detect-Goal Move-to-Goal Σ actuators

38 Visualization of Vector Field for Ex. 5 Avoid-obstacle V magnitude = 0 S d G S R for for for S = obstacle s sphere of influence R = radius of the obstacle G = gain D = distance robot to obstacle s center V direction = radially along a line between robot and obst. center, directed away from the obstacle R d < d > d S R S Obstacle Obstacle

39 Visualization of Vector Field for Ex. 5 Move-to-goal Output = vector = (r,φ) (magnitude, direction) V magnitude = fixed gain value Goal V direction = towards perceived goal

40 Visualization of Vector field for Ex. 5 Avoid-obstacle + move-to-goal Linear superposition (vectorial weighted sum) O O G

41 Ex. 5: Issue with Motor Schemas Detect-obstacles Avoid-obstacle sensors Detect-Goal Move-to-Goal Σ actuators Generate-direction Noise For avoiding to get stuck in local minima of the vector field (typical problem of vector field approaches)

42 Evaluation of Motor Schemas + Support for parallelism: motor schemas are naturally parallelizable + Run time flexibility: schemas = software agents -> reconfigurable on the flight - Robustness -> well-known problems of potential field approach -> extra introduction of noise (not clear method for exploiting that generated by sensors, ) - Slow and computationally expensive sometimes - No HW retargetability: do not provide HW compilers; do not take into account the system as a whole

43 Evaluation of both Architectures in Practice In pratice (my expertise) you tend to mix both and even more The way to combine basic behavior (collaborative and/or competitive) depends on how you developed the basic behaviors (or motor schemas), reaction time required, on-board computational capabilities, Pierre Arnaud s work (thesis and book EPFL, 2000, see references at the end); Masoud Asadpour s work (thesis EPFL, 2006, see reference at the end) went in this direction for different reasons

44 Conclusion

45 Take Home Messages Hardware design choices are at the basis of embedded systems (typical building blocks can be modules or off-the-shelf components) Good understanding of these resources is key for software-controlled resource management Perception-to-action loop is key in robotics, different complexity and level of control architectures A given behavior can be obtained with different control architectures Control architectures are characterized by more or less parameters which are often not so easy to tune

46 Additional Literature Week 10 Books Braitenberg V., Vehicles: Experiments in Synthetic Psychology, MIT Press, Siegwart R. and Nourbakhsh I. R., Introduction to Autonomous Mobile Robots, MIT Press, Arkin R. C., Behavior-Based Robotics. MIT Press, Everett, H. R., Sensors for Mobile Robots, Theory and Application, A. K. Peters, Ltd., 1995 Arnaud P., Des moutons et des robots, Presses Polytechniques et Universitaires Romandes, PhD Theses Asadpour M., Behavior Design in Microrobots: Hierarchical Reinforcement Learning under Resource Constraints, EPFL PhD Thesis Nr. 3682, Lausanne, Switzerland, November 2006.