INTEGRATION AND EVALUATION OF SENSOR MODALITIES FOR POLAR RO(3)

Upon selecting the suite of sensors,each sensor was integrated into the PRISM polar robot.A software API was developed in Java.Drivers were written for each sensor based on the API.Fusion of sensor data was also performed to create robust virtual sensors.Sensor connectivity was also addressed such that multiple sensors could be connected to a single computer.Finally,the task of

6

physically mounting each sensor at the point where it would be most a?ective was addressed.

Once the sensors were integrated into the PRISM rover,series of experiments were designed and performed to evaluate the sensor suite’s performance.These experiments were performed both locally at the University of Kansas and on Greenland’s ice cap.These experiments were designed to verify important sensor speci?cations,determine the a?ects of locality on sensor performance, and determine any potential weaknesses that must be addressed for future?eld experiments.

1.3Thesis Structure

This thesis is pided into10chapters.Chapter2presents the background knowledge relative to this work.In particular,three major areas of robotic sensing: navigation and localization,collision detection and avoidance,and proprioception are presented.Sensor fusion is also discussed.Chapter3discusses related work in the areas of planetary rovers and polar rovers.Chapter4presents the sensing requirements and selection criteria used for developing the PRISM rover.Chapter 5describes the entire sensor suite for the PRISM rover and discusses how each sensor relates to the requirements and selection criteria.Chapter6discusses how the sensors were integrated into the PRISM rover.Integration was pided into three areas of mounting,connectivity,and software.

Chapters7,8,and9discuss the validation of the sensor suite.Chapter7 discusses the evaluation goals and provides an overview of experiments performed locally and during the2003and2004Greenland?eld seasons.Chapter8discusses speci?c experiments that were performed to validate the sensors.Chapter9 presents the results and their analysis.

Chapter10concludes this thesis,presenting the contributions,limitations, and the future work.

7

2.BACKGROUND

This research focuses on the selection,integration,and validation of sensors for polar robots.Many sensor modalities exist to support autonomous and semi-autonomous mobile robots.The choice of sensors to integrate is problem-dependent.

The desired level of autonomy is a major factor in?uencing the sensor selection.A fully autonomous mobile robot must be capable of sensing its environment,interpreting the sensor data,and carrying out actions to achieve its goals.Vehicles that are teleoperated have no on-board autonomy and rely on commands from a remotely situated operator.Vehicles with supervisory control are between these extremes.They have some autonomy;however,their state is monitored remotely.A remote operator is capable of assigning new tasks, providing guidance,or temporarily taking control of the robot.The amount of control given to an outside controller is mission dependent[22].

The PRISM project’s mobile robot is semi-autonomous.It operates autonomously once a set of goals is assigned.The goals can be formulated by the rover’s own control system,or assigned by a human operator.A remote operator is also capable of overriding the autonomous control of the vehicle.To select sensors that facilitate the needs of the PRISM polar robot,sensors for mobile robots were investigated,along with the topic of sensor data fusion.

2.1Navigation and Localization

A semi-autonomous mobile robot must be capable of traveling from one point to another point.Within its current environment,the mobile robot must be capable of determining its location relative to the target destination(localization).It

8

must also be capable of determining the proper change in trajectory such that it can travel toward the destination.The sensing requirements for navigation and localization can be classi?ed into two categories:position and heading/orientation.

2.1.1Position Sensors

Determining the mobile robot’s current position plays a dual role for the PRISM project.Intelligent navigation requires knowledge of the rover’s location within the current world model.Location information could involve the precise location on a local or global map,or it could involve the distance traveled since its movement.

The robot’s position also plays a role in the scienti?c objectives of the PRISM project.Position information is shared both with the radar and the intelligent systems.Precise positioning is necessary to form an accurate radar image.

Position is determined either locally or globally.Locally,relative positioning determines the position of the bot within a local region relative to some?xed point within that region.Globally,absolute positioning determines the robot’s precise location with respect to a geodetic world model.A geodetic world model is a representation of position on a coordinate system that represents the world. For instance,Latitude,Longitude,and Altitude may be utilized.The Universal Transverse Mercator(UTM)coordinate system is another common geodetic world model that pides the Earth’s surface in to grids where position is represented as x,y,z coordinates within these grids.Relative positioning satis?es the requirements of the PRISM project.

In this section,various positioning sensors will be discussed with focus on how these sensors may be used for relative positioning.The bene?ts and limitations of each sensor are also described.

9

2.1.1.1Dead Reckoning

Dead reckoning approximates the position of the mobile robot by measuring its rate of motion and determining the relative change in its position.Dead reckoning utilizes various devices such as shaft encoders,Doppler,and accelerometers.

Shaft or quadrature encoders are some of the least expensive and simplest devices used for dead reckoning.These sensors measure the rotational rate of the vehicle’s wheels or axles.Given these rates,the change in position and possibly orientation of the robot can be determined.Incremental encoders often use a photosensor to calculate the number of alternations of a pattern on a disk which rotates about the axis.Wheel slippage often results in error accumulation over time[31].

Doppler technology has been used for dead reckoning.Doppler-based sensors measure the time-of-?ight of ultrasonic or microwave bursts to determine the velocity of the 7d21773343323968011c92dfing this velocity,it is possible to approximate the vehicle’s position over time.Doppler is however susceptible to interference.Its accuracy is also reduced on rough terrain where sudden changes in elevation exist [20].

Accelerometers can also be employed.The vehicle’s acceleration must be doubly integrated in order to approximate the vehicle’s position.Similar to the Doppler technology,this sensor does not work well on rough terrain.The sudden impacts from the bumpy road will add noise to the acceleration measurements [20].

Dead reckoning provides decent estimates,but lacks accuracy.Error accumulation may become too great,and redundancy can likely improve accuracy. For applications requiring precise navigation,dead reckoning is infeasible.

10