ACCURACY IMPROVEMENT FOR PHYSICAL ROBOT Gal Lerman Dorin

ACCURACY IMPROVEMENT FOR PHYSICAL ROBOT Gal Lerman, Dorin Ben-Zaken

The Project The goal is to improve the accuracy of a robot. Based on a Motion Planning workshop for physical robots. Software improvements to the robot driver. Additional sensors supported by extra hardware.

The Workshop We have a computer controlled robot. The robot maneuvers in an obstacle filled room. The robot starts at a certain position and must reach a target position.

Our robot…

Just kidding…The i. Robot Create Two wheel differential drive robot. Based on the well known Roomba vacuum cleaner.

The i. Robot create Left and right wheels are controlled separately. Built-in sensors: � � Left and right bumpers (know when the robot collides). Wheel encoders (measures the number of wheel revolutions). Lacking a high-quality distance sensor…So we added our own.

Encoder Inaccuracy Reasons Wheel slippage counts as movement Finite encoder resolution results in round-up errors Errors are accumulated over distance – the greater the distance the bigger the error

Workshop Architecture Distance sensor Analog out Bluetooth RS 232 BAM

Robot Driver Communication between the robot and the PC is done by serial communication over Bluetooth. The robot driver sends command packets complying to the robot’s protocol. The robot sends packets containing sensor data every 15 ms. Also implemented a simulator driver (using a shared interface).

Scene Loader Scenes contain information about: � The bounding box of the room. � Obstacles. � Robot start and goal positions. Scenes are loaded from files.

Scene Solver A general interface for solving the motion planning problem. Uses the PRM (Probabilistic Roadmap) algorithm: � Samples random free positions in the room. � Builds a graph by connecting near positions. � Reduces the problem to a shortest-path in graph problem.

Graphical User Interface (GUI) A user interface for interacting with the system. Also shows sensor data, roadmap and more. User Interface Scene Loader Robot Mover Scene Solver Interface PRM Scene Solver Robot Driver Interface Physical Robot Driver Simulator Robot Driver

Dynamic Obstacles We added a remote-controlled Roomba to act as a dynamic obstacle. Distance sensor is used to detect the Roomba. Requires algorithm modifications: � � � When detecting a dynamic obstacle we add a temporary static obstacle. Update the roadmap graph so that colliding edges are removed. Re-plan using the new roadmap

Robot Competition Two PCs controlling different robots. Another PC acting as referee and coordinates between them. First robot to reach its destination wins. Communication over TCP.

Lack of accuracy demonstration video

The Project - Hardware

Original Project Architecture XBee module Zigbee Communication XBee on USB dongle RS 232 Arduino microprocessor Gyroscope Accelerometer RS 232 i. Robot create PC running robot’s motion path software

Original Project Architecture - limitation Robot response to command isn’t reliable - Robot freezes a lot Other developers in the community complained about Xbee reliability in full duplex high packet rates scenarios i. Robot Command Module is a microcontroller designed for i. Robot Create, in hind sight perhaps it was better to use it instead of Arduino

Project Architecture Based on Workshop i. Robot create RS 232 BAM Bluetooth PC running robot’s motion path software Analog out D/A Digital out * Arduino microprocessor Gyroscope Accelerometer Analog out Distance sensor (*) Here we use Time Division Multiplexing on the 3 sensor data streams

Project Architecture Based on Workshop - cont Xbee free architecture Reuse of BAM as proven communication channel between robot and pc Ardunio interfaces with the sensors and converts their data to n digital bits D/A converters Arduino digital output to robot’s analog input Time Division Multiplexing to transfer 3 sensor data stream using 1 analog input in the robot Full use of the robot’s 4 digital inputs � 2 name bits – naming the sensor data � Start, Stop bits to solve Arduino-Robot synchronization problem (packet containing gyro’s name with distance data)

Project Architecture Based on Workshop - limitation Only 8 bit D/A available in TAU where we needed at least 9 (8 bits for angle [0, 180] and 1 bit for sign) Complex architecture A simpler idea came to mind…

Final Project Architecture Gyroscope XBee on USB dongle Bluetooth Arduino microprocessor XBee Bluetooth Sensor data Power Commands i. Robot create Analog out Distance sensor PC running robot’s Gyro packets RS 232 BAM motion path software

Final Project Architecture - cont PC has 2 serial connections � Xbee channel from Arduino � BAM channel to/from Robot Xbee back in action but in only one sided communication and only sensor data to PC Reuse of BAM as proven communication channel between robot and PC

Accelerometer limitation Double integration to get location data from acceleration causes massive accumulation of errors Tilt of the sensor is interpreted as movement Mobile Robot Positioning – Sensors and Techniques by J. Borenstein, H. R. Everett, L. Feng and D. Wehe.

Software improvements Wait angle command � Specify desired degrees and let the robot stop after executed � Requires to stop the sensor stream before command is lunched and resume it after command executed Combine angle reading from robot’s odometry and gyroscope to get a more accurate angle

UMBmark – Odometry calibration Based on a University of Michigan article from 1994 Designed to correct systematic odometry errors Systematic errors are caused by Inaccurate wheel base � Inaccurate wheel radius � Calculates alpha and beta α - the error in degrees when the robot tries to turn 90 degrees � β – the angle in degrees the robot has veered off a straight path after running the straight line �

UMBmark – Odometry calibration cont Correct α by turn. Angle = requested. Turn. Angle*(90/(90 -α)) Correct β by driving in an arc instead of straight line. Arc with radius of R = (L/2)/ β ) We couldn’t use it because of drive-with-radius command is limited to only 2 m radius

Results

Results - cont

Results - cont video

Conclusions Wait angle accounts for most of the accuracy improvement No distance accuracy improvement because of accelerometer sensor’s limitations Possibly better to use a laser sensor instead of accelerometer Wait distance would probably improve the distance accuracy greatly, but at the cost of loosing response to dynamic obstacles

Thank you Questions?