Navigational System For An Autonomouse Farming Vehicle Group 942

Project Topic Navigational System for an Autonomous Farming Vehicle

Group GPS 942 ► Frank K Ocran – Introduction ► Darius Plausinaitis – System setup ► Saulius Pusinskas – Test results ► Jon Johansen Wedel – Improvements ► Ramunas Dirmeikis – Improvements and conclusions

Introduction

Trends in autonomous vehicles ► Autonomous vehicles are enabled by advances in - vehicle concept and mechanical design - vehicle concept and mechanical design - sensors (eg GPS) - sensors (eg GPS) - vehicle electronics (vectronics) - vehicle electronics (vectronics) - control and planning - control and planning - user interface - user interface

Recent trends in GPS Technology ► Removal of SA ► Satellite Based Augmentation System such as WAAS, EGNOS ► Carrier phase DGPS leading to millimeter accuracy ► Third frequency L5

Why autonomous farming vehicles? ► Decreased labour costs ► Increased machinery operation times ► Possibility of using smaller farm engines ► High efficiency - less damage crops due to over fertilization - less damage crops due to over fertilization - No overlapping during seeding, spraying etc - No overlapping during seeding, spraying etc

John Deere 5510

8200 Autonomous Agriculture Tractor Autonomous Agriculture Tractor.

Autonomous Orchard Sprayer

Problem Statement ► To design and implement part of a navigational system for an autonomous farming vehicle ► System will output the position and heading of vehicle using RTK algorithms with 3 GPS receivers in DGPS set up.

Limitation of the Project ► Receivers used was of Topcon and Ashtec manufacture and radio transceivers used was also of Satel manufacture. ► Test was not carried out using an actual farming vehicle. ► Main focus was the implementation of RTK positioning. ► Steering of the vehicle, digital mapping of the fields and roads for vehicle to navigate ► Other sensory input systems

System Requirements ► Top operating straight line speed is 10m/s ► Distance between two on board receivers is at least 2m ► Variance of the position of any of the on-board receivers is less than 10cm ► Variance of the direction of vehicle is less than 2 degrees ► Minimum update rate of GPS receiver was 5Hz

Method

Why this method? ► An efficient way ► Easy to control and monitor the problems ► Easy to control the work force ► Preliminary results are obtained quickly

System Setup

a b h GPS antenna “Rover 2” GPS antenna “Rover 1” Reference GPS antenna ► What is needed to navigate an autonomous vehicle?  Position of the vehicle  Heading of the vehicle

System interface Processing unit ► System interface ► Positioning ► Filtering Radio modem GPS Main engine Filtering Positioning

System Interface

► Main engine  Purpose: data mining and data exchange between all functions and presentation of the results  Communication ► The problem: to transfer data from the reference receiver ► The solution: to use two radio modems  Problem: Matlab is too slow  Solution: a C code is used for data mining

System Interface: Conclusions ► The C code is able to record the data at the 5Hz rate or even faster ► The available radio links are not always stable, especially when the vehicle is moving ► The downside: calculations must be done in the post processing mode

Positioning

Positioning ► The problem: To calculate position and heading of the vehicle with sufficient accuracy ► The solution: To use Differential GPS to obtain desired accuracy

Positioning: How The DGPS Works Two receivers (i,j) – two satellites (k,l): Ionosphere k ij l The single differences: The double difference: Troposphere

Positioning: Tests And Results

Filtering ► The problem: Some of the errors are random and are not removed by the DGPS ► The solution: To use a Kalman filter

The Kalman Filter ► Problem: How to describe and predict the trajectory of the vehicle? ► Solution1: let’s use vehicles velocity and curvature of the trajectory ► Solution2: let’s extrapolate next point on the basis of the last three points

Filtering: Tests and Results

Final Tests

The System To Test GPS receivers, radio transceiver, PU Main engine Positioning algorithm Kalman filtering Hardware layer Main engine layer Positioning layer

Test Session 1 - Setup ► Static tests with one reference and one rover receiver ► Short time span (1 minute) measurements

Drift Of The Baseline Coordinates

Sudden large jumps of coordinates

Test session 1 – errors discovered GPS receivers, radio transceiver, PU Main engine Positioning algorithm Kalman filtering Hardware layer Main engine layer Positioning layer

Test session 1 - conclusions The performance of the system is unacceptable The performance of the system is unacceptable All the tests should be redone after finding and removing the error source All the tests should be redone after finding and removing the error source Version control should be used for the code Version control should be used for the code Pinpointing the errors took substantial time

Test session 2 - setup ► Static test ► Kinematic test

Unacceptable positional accuracy

Coordinate transformation problem

Unsatisfactory Kalman filter performance

Positioning performance against time

Test session 2 - conclusions ► The system developed is not able to perform within the specifications ► The computational time satisfies the requirements

Final Tests - Discussion ► Complex system needs complex treatment ► Hardware and software itself should be tested prior to using it in the system tests ► Longer time span measurements should have been done

Improvements and Conclusion

Improvements ► What has been done since project submission? ► The Kalman filter ► The DGPS function

The Kalman filter ► Improvements proposed  Include more states in the filter  Include limitations in the filter  Find alternative approaches

Principle of the improved filter

The DGPS function ► Include Doppler measurements ► Solve integer ambiguities ► Use more epochs in the solution ► Use information about relative position of rovers

LAMBDA Reintroduction of the LAMBDA method

Standard Deviation Avg: m Standard Deviation R1: m Standard Deviation R2: m Standard Deviation Avg: m Standard Deviation R1: m Standard Deviation R2: m

Memory Use of the previous epoch for position estimation

Standard Deviation Avg: m Standard Deviation R1: m Standard Deviation R2: m Standard Deviation Avg: m Standard Deviation R1: m Standard Deviation R2: m

Memory and LAMBDA Combining more epochs with the LAMBDA method

Standard Deviation Avg: m Standard Deviation R1: m Standard Deviation R2: m Standard Deviation Avg: m Standard Deviation R1: m Standard Deviation R2: m

Yang, Goad and Schaffrin method Another way of estimating integer ambiguities

Yang, Goad and Schaffrin method ρ L1 = r + ε ρL1 ρ L2 = r + ε ρL2 Φ L1 = r + λN L1 + ε ΦL1 Φ L1 = r + λN L1 + ε ΦL1 Φ L2 = r + λN L2 + ε ΦL2 Φ L2 = r + λN L2 + ε ΦL2Where: ρ L1, ρ L2 - double differenced code measurements Φ L1, Φ L2 - double differenced phase measurements r - double differenced geometric range r - double differenced geometric range

Yang, Goad and Schaffrin method Matrix notation :

Standard Deviation Avg: m Standard Deviation R1: m Standard Deviation R2: m Standard Deviation Avg: m Standard Deviation R1: m Standard Deviation R2: m

Conclusions on the improvements ► Position algorithm  Move from search techniques  Increase epoch number  Kalman filter

Conclusions on the improvements ► Position algorithm ► Filtering algorithm  New approach

Conclusions on the improvements ► Position algorithm ► Filtering algorithm ► Main Engine  Real Time  DSP

Conclusions on the improvements ► Position algorithm ► Filtering algorithm ► Main Engine ► Hardware  Radio link  GPS receivers

Final Conclusions

The End Thank you for your atention