1
Behavior Based Robotics: A Wall Following Behavior Arun Mahendra - Dept. of Math, Physics & Engineering, Tarleton State University Mentor: Dr. Mircea Agapie CONCLUSIONS Our work demonstrates successful integration of Artificial Intelligence and Systems Theory algorithms. Due to real-life constraints, behavior of robot in simulation and real-life can be very different. FUTURE WORK Explore the effect of changes in PID control parameters, especially in regard to differences between simulated and real-life behavior. Develop a memory-based autonomous learning behavior. The robot will learn the values of the PID parameters without supervision. Coordinate behaviors of multiple robots to explore in teams. For additional information please contact: Mircea Agapie Dept. of Math, Physics & Engineering Tarleton State University [email protected] Collision Avoid Behavior The CollisionAvoid behavior is activated when the reading from any sensor is less than the minimum threshold value. This is how the robot determines if an object is close enough for a collision. When an object is detected too close to the robot, it avoids a collision by moving away from it in the opposite direction. CollisionAvoid has the highest priority and Therefore, it can override other behaviors. Explore Behavior Proportional-Integral-Derivative Controller Experimental values for PID control loop multipliers Proportional Gain: k P = 0.15 Integral Gain: k i = 0.15 Derivative Gain: k d = 0.15 AmigoBot The robot communicates with the computer across a (802.11) wireless network. Information packets are sent every 100 milliseconds. Wall Follow Behavior Abstract One of the most important areas of research in robotics is the emergence of complex behaviors from simple ones. A robotic behavior of level of complexity N is designed in two steps: first it is decomposed into modules, which are simpler behaviors, on level N-1. Then a control algorithm is designed, which decides which lower-level behavior is active at any given time. We develop a “wall-following” behavior for the ActivMedia AmigoBot according to the general process described above. The novelty of our work is twofold: we mix traditional, threshold-based behaviors with behaviors “borrowed” from systems theory (the PID controller) and we include history as another input to the control algorithm. This provides a solution to the classical problem of aliasing, and gives robustness to the emergent behavior, as proved by testing in various scenarios, using a simulator as well as the real robot. Grouping Sensors to form Eyes Flow of Control Ultrasonic Sonar Am igoBot Eight sonar units built into the robot constitute the input sensors. The robot identifies external obstacles and wall-to-follow by processing inputs from these sensors. When these sensors return a reading, it is processed and as a reaction, a state transition in the system occurs. Different states reflect different reactive behaviors. Simulation of the robot taking Convex Turns Simulation of the robot taking Concave Turns P is proportional to the error. I is proportional to the cumulative history of the error. D is the predicted error. Simulation of the robot exploring an empty room to find a wall. The Explore behavior is activated when there are no objects near by. If the robot senses any object, it will change its path and move towards the object. Sensors and Positions Experimentally, a translational speed of 200 mm/sec was found ideal for navigating without colliding with objects. Distance traveled @ 200 mm/sec before a packet arrives = 200 *.1 = 20mm. The “blind distance” covered in this case is only 20mm, which is very small. Implementation of the Eye using C++ Amigobot is a member of the Pioneer family of mobile robots manufactured by ActiveMedia Robotics.

Behavior Based Robotics: A Wall Following Behavior Arun Mahendra - Dept. of Math, Physics & Engineering, Tarleton State University Mentor: Dr. Mircea Agapie

Embed Size (px)

Citation preview

Page 1: Behavior Based Robotics: A Wall Following Behavior Arun Mahendra - Dept. of Math, Physics & Engineering, Tarleton State University Mentor: Dr. Mircea Agapie

Behavior Based Robotics: A Wall Following Behavior

Arun Mahendra - Dept. of Math, Physics & Engineering, Tarleton State University

Mentor: Dr. Mircea Agapie

CONCLUSIONS

Our work demonstrates successful integration of Artificial Intelligence and Systems Theory algorithms.

Due to real-life constraints, behavior of robot in simulation and real-life can be very different.

FUTURE WORK

Explore the effect of changes in PID control parameters, especially in regard to differences between simulated and real-life behavior.

Develop a memory-based autonomous learning behavior. The robot will learn the values of the PID parameters without supervision.

Coordinate behaviors of multiple robots to explore in teams.

For additional information please contact:Mircea AgapieDept. of Math, Physics & EngineeringTarleton State [email protected]

Collision Avoid Behavior

The CollisionAvoid behavior is activated when the reading from any sensor

is less than the minimum threshold value. This is how the robot

determines if an object is close enough for a collision. When an object is

detected too close to the robot, it avoids a collision by moving away from it

in the opposite direction. CollisionAvoid has the highest priority and

Therefore, it can override other behaviors.

Explore Behavior

Proportional-Integral-Derivative Controller

Experimental values for PID control loop multipliers

Proportional Gain: kP = 0.15

Integral Gain: ki = 0.15

Derivative Gain: kd = 0.15

AmigoBot ¶

The robot communicates with the computer across a (802.11) wireless network.

Information packets are sent every 100 milliseconds.

Wall Follow BehaviorAbstractOne of the most important areas of research in robotics is the emergence of

complex behaviors from simple ones. A robotic behavior of level of

complexity N is designed in two steps: first it is decomposed into modules,

which are simpler behaviors, on level N-1. Then a control algorithm is designed,

which decides which lower-level behavior is active at any given time. We

develop a “wall-following” behavior for the ActivMedia AmigoBot according to

the general process described above. The novelty of our work is twofold: we mix

traditional, threshold-based behaviors with behaviors “borrowed” from systems

theory (the PID controller) and we include history as another input to the control

algorithm. This provides a solution to the classical problem of aliasing, and

gives robustness to the emergent behavior, as proved by testing in various

scenarios, using a simulator as well as the real robot.

Grouping Sensors to form Eyes

Flow of Control

Ultrasonic Sonar

AmigoBot

Eight sonar units built into the robot constitute the input sensors. The robot identifies external obstacles and wall-to-follow by processing inputs from these sensors.

When these sensors return a reading, it is processed and as a reaction, a state

transition in the system occurs. Different states reflect different reactive

behaviors.

Simulation of the robot taking Convex Turns

Simulation of the robot taking Concave Turns

• P is proportional to the error.

• I is proportional to the cumulative history of the error.

• D is the predicted error.

Simulation of the robot exploring an empty room to find a wall.

The Explore behavior is activated when there are no objects near by. If the

robot senses any object, it will change its path and move towards the object.

Sensors and Positions

Experimentally, a translational speed of 200 mm/sec was found ideal for

navigating without colliding with objects.

Distance traveled @ 200 mm/sec before a packet arrives = 200 *.1 = 20mm.

The “blind distance” covered in this case is only 20mm, which is very small.

Implementation of the Eye using C++

¶ Amigobot is a member of the Pioneer family of mobile robots

manufactured by ActiveMedia Robotics.