Heterogeneous Teams of Modular Robots for Mapping and Exploration - PowerPoint PPT Presentation

heterogeneous teams of modular robots for mapping and exploration l.
Skip this Video
Loading SlideShow in 5 Seconds..
Heterogeneous Teams of Modular Robots for Mapping and Exploration PowerPoint Presentation
Download Presentation
Heterogeneous Teams of Modular Robots for Mapping and Exploration

play fullscreen
1 / 40
Download Presentation
Heterogeneous Teams of Modular Robots for Mapping and Exploration
Download Presentation

Heterogeneous Teams of Modular Robots for Mapping and Exploration

- - - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript

  1. Heterogeneous Teams of Modular Robots for Mapping and Exploration Speaker: Hyokyeong Lee Feb 13, 2001

  2. Abstract • Design of a team of • Heterogeneous and centimeter-scale robots • Collaborate to map and explore unknown environments • Millibots are configured from modular components • Sonar and IR sensors,Camera,Communication,Computation,Mobility • For mapping and exploration • Critical to know the relative position of each robot • Novel localization system • Sonar-based distance measurements to determine the positions of all the robots in the group • Occupancy grid Bayesian mapping algorithm to combine the sensor data from multiple robots

  3. Introduction • Advantages of a team of robots • Load distribution • Smaller, lighter, less expensive robot • Sensing • A team of robot • Perceive environment from multiple disparate viewpoints • Task completion by a team of collaborating robots • A single robot • A single point of view • Distributed viewpoints • Surveillance, monitoring,demining and plume detection

  4. Introduction • Traditional robots • Designed with broad array of capabilities • Redundant components to avoid system failure • Large,complex,expensive system • Robot teams • “Build simple inexpensive robots with limited capabilities that can accomplish the task reliably through cooperation” • Each robot may not be very capable, but team accomplishes useful task • Less expensive robots that are easier to maintain and debug • Each robot is expendable, reliability in numbers

  5. Introduction • Size of a robot determines its capabilities • Hierarchical robot team (Figure 1) • All Terrain Vehicles(ATVs) • Pioneer robots • Medium-sized Tank robots • Centimeter scale Millibots

  6. Introduction

  7. Introduction • All Terrain Vehicles(ATVs) • Range of up to 100 miles • Completely autonomous • Extensive computational power • Transport and deploy groups of smaller robots • Pioneer robots • Port-based adaptive agents • Allow the team to dynamically exchange algorithm and state information while on-line

  8. Introduction • Medium-sized Tank robots • Medium-sized, autonomous robots with infrared and sonar arrays • Swivel-mounted camera • On-board 486 computer • Capable of individual mission • Serve as the leader and coordinator for a team of Millibots

  9. Introduction • Millibots • Small and lightweight robots • Maneuver through small openings and into tight corner to observe areas that are not accessible to the larger robots • Small so less noticeable

  10. Milibots • Size • Primary factor that determines what a robot can do and where it can go • Small robot • Advantage • Crawl through pipes, inspect collapsed buildings, hide in small inconspicuous spaces,dramatic impacts for surveillance and exploration task • Disadvantage • Limited mobility range, limited energy availability, reduced sensing, communication and computation ability due to size and power constraints • Trade-off

  11. Small Robots • Khepera • Ants • FIRA&RobotCup

  12. Khepera • Small size & computing complexity • Significant on-board processing • Modular • Support addition of sensor and processing modules • Work alone or communicate and act with other robots • Lack a significant feature: self-localization • Operate in an unknown environment • Combine sensor information • Act as a central, cohesive unit • Rely on fixed position global sensor or internal dead-reckoning • Ineffective as a deployable set of robots • A pair of centimeter sized wheels • Restricts the robot’s clearance to about 3mm

  13. Ants • Same scale as the Millibots • Designed to be used in groups or teams • Limited in sensing • Designed to explore reactive social behaviors • Not support a real-time communication link • Not exchange information necessary to produce maps or models of the environment • Built with a fixed architecture to achieve scale • Propulsion, sensing, processing are combined • Addition of new functionality requires complete redesign • Inability to localize • Rely on strong light source for orientation and encoders for dead reckoning • Without a means for determining position, little context in which to evaluate data

  14. FIRA&RobotCup • Small-scale cooperating robots • Coordinate to perform complex actions against a coordinated attack • Extremely limited in sensing capabilities • Position sensing via a global camera • Little or no sensors on the robots themselves • Blind and unable to respond to real world event without external camera

  15. Specialization • Specialization • Achieved by exploiting the nature of a heterogeneous team • Instead of all-equipped robot, build specialized robots for a particular aspect of the task • Scenarios • Robot team composed of robots with various range and position sensors but only limited computation capabilities • Omitting the unnecessary capabilities • Reduction of power, volume, and weight of the robot • Disadvantage • Many different robots need to be available to address the specific requirements of a given task

  16. Modular Architecture • Specialization through modularity optimizes resource • A robot with only mission specific modules • Minimum size and cost of the robot • Reduction of unnecessary payload • Less weight, less power • Smaller and cheaper • Robots in large numbers to achieve dense sensing coverage, team level adaptability, and fault tolerance • Figure 2 • Coordination between modules • Dedicated slots • Fixed connections that support up to six sensors or actuator modules • Common I2C bus • High speed, synchronous clock • Two-way data communication line

  17. Millibot’s architecture and subsystems

  18. Millibot subsystems • Seven subsystems • Main processor module • Communication module • IR obstacle detection module • Two types of sonar modules • Motor control module • Localization module

  19. Communication • Essential in a coordinated team • Collaborative mapping and exploration • Detailed and abstract information • Not easily conveyed implicitly • To provide two-way communications within group • Each robot is equipped with radio frequency transmitter and receiver • Exchange data at 4800 bps at a distance of up to 100 meters • Units based on size and power considerations

  20. Sensors • A set of ultrasonic sonar modules • Short-range distance information • For obstacles between 0 and 0.5m • Ideal for work in tight or cluttered areas • Long-range distance information • For obstacles between 0.15m and 1.8m • Effective in hallways or open office space • Potential complication with ultrasonic based sensor • Interference with similar modules on other robots • Sonar elements operate at a fixed frequency determined by mechanical construction • Carry infrared proximity module to overcome this problem • Provide an array of five tunable, infrared emitter-detector pairs

  21. Sensors • Still problem in spite of sensor • Anomalies in real situation • Need high bandwidth information during a mission for analysis by a higher level process or operator • Camera module as a solution • External mini camera • A small video transmitter • included with the module to transmit the raw video signal to an external processor or remote viewing station • Power circuitry • Allows the camera and its transmitter to be switched on and off via control signal from the Millibot • Resources minimization • Only one receiving station and associated monitoring device is needed per Millibot group • Cannot be used continuously due to limited size of the battery

  22. Collaboration • To know relative position and orientation of the robots • Localization method • Dead reckoning • Accuracy problems due to integration error and wheel slippage • Camera-based localization • Not feasible in small robots • Global Positioning Systems (GPS) • Not appropriate for use in small robots that operate mostly indoors • Landmark recognition, map-based positioning • Require excessive local computational power and Sensing accuracy on Millibots

  23. Collaborative Localization • Millibot localization system • Based on trilateration • Determines the position of each robot based on distance measurements to stationary robots with known positions • Localization module • Utilizes ultrasound and radio pulses • Act as both emitter and receiver • Figure 3 • Low-cost ultrasonic transducer • To produce and detect beacon signals • Coverage of 360 degree, measure distance up to 3m with a resolution of 8mm while consuming 25mW • Paramount in achieving a localization system at this scale

  24. 343m/s 3X108m/s

  25. Localization Algorithm • Maximum likelihood estimator • Measurements are noisy and missing • Purely geometric approach is over-constrained, does not yield a solution • Assumption • We know the position of an orientation, (x0,y0,0) of all the robots at time t0 • How determine the position, (x1,y1,1), of the robots at time t1 after they have moved • Estimate the new position based on informaton • Dead reckoning • Distance measurements

  26. Dead reckoning • Millibots always move according to “vector commands”, (,d) • :rotation in place over an angle  • d:forward straight-line motion over a distance d • : angle over which the robot rotates while moving forward • Unplanned rotation due to wheel slippage and calibration errors in the controller

  27. Dead reckoning

  28. Distance measurements • Each robot that moved pings its localization beacon to determine its distance to all the other robots • Resulting measurement data provides accurate data to overcome the drift encountered in localization algorithm based on dead reckoning alone

  29. Likelihood • Assume that dead reckoning data and distance measurements are normally distributed • Dead reckoning • The likelihood that a robot moved over • Over an angle • Distance • given initial position • final position

  30. Likelihood • Distance measurements • The likelihood that the measured distance between two robots i,j is equal to Di,j

  31. Likelihood • Total conditional likelihood function is the product of all the conditional likelihoods introduced • The most likely robot positions are found by maximizing Ptot with respect to the new robot positions

  32. Implementation Issues • Optimization of the conditional probability density function • Formulated as a weighted nonlinear least-square problem • BFGS nonlinear optimization algorithm • When no prior information about robot positions is available, the BFGS algorithm may get stuck in a local minimum • Dead reckoning data provides a good starting point • Only a few iterations are necessary to reach optimum • Taking the best-out-of-five randomly initialized runs never fails to find the global optimum • Filter the raw measurement data • Obtain good results with the above algorithm • Improvement of accuracy of the algorithm by using more than three robot beacons • Median and mean filtering reduced the standard deviation of the distance measurement

  33. Mapping and Exploration • A group of Millibots can be equipped with similar sensors to cover more area in less time than a single robot • Team leader(or human operator) • Utilize the robot’s local map information • Direct the Millibot around obstacles • Investigate anomalies • Generate new paths • Merge the information from several local maps into a single global map • Map aids in path planning for the movement and positioning of the team during exploration

  34. Mapping and Exploration • Occupancy grid with a Bayesian update rule • Produce maps of the environment • Allows the combination of sensor readings from different robots and different time instances • Occupancy value • 1: occupied by an obstacle • 0: free cell

  35. Experimental Results • Task to explore and map as much area as possible before the team failed • Possible failures included • Loss of localization, loss of battery power, loss of communications • For each experiment • Three Millibots equipped with sonar arrays for collecting map information • Two Millibots equipped with camera modules to aid in obstacle identification and provide a level of fault tolerance • All equipped with localization module • CyberRAVE • A central control server and a set of distributed GUI • Operator directs the robots by setting goals,querying maps, and viewing live sensor data

  36. Operation of the Team • First experiment • Test and verify the team’s ability to localize and collect map data • Second experiment • At the center right of the hallway which included a cluster of objects against one of the walls • Detect and avoid the obstacles and remained operational for more than an hour • Loss of a camera robot but mission was continued • Third experiment • A large number of obstacles which were small and low to the ground making them invisible to the sonar sensors • Camera modules played a significant role • Prior to moving any robot , camera scan the area in front of the robot • Reduced the exploration speed

  37. Merging result

  38. Metrics • Evaluate and compare the performance of the team

  39. Summary • Present the design of a distributed robotic system consisting of very small mobile robots • Modular fashion to expand the capabilities • Novel ultrasound-based localization system • Does not need require any fixed beacons • Using Millibots alternatively as beacons and as localization receivers • Can reposition while maintaining accurate localization estimates at all times

  40. Critics • Ambiguous in path planning by the operator • By intuition? • By which ground, can we say that a deviation of 3% in the distance measurement is good? • Is 3% acceptable? • In occupancy grid, how is the homogeneous cell size determined? • Suggestion • How about using learning and reasoning algorithm to implement fully autonomous system