Robotic Raspberry Pi powered lawn mower

Last week I got my second Raspberry Pi. If you don’t know what it is already, it’s a 25$ fully blown computer with two USB ports, ethernet, video, audio and HDMI port of credit card dimensions. It has sixteen programmable GPIO ports, external display and camera port and is powered by single micro-usb. It’s power consumption is 3,5 Watts and it’s capable of HD video output. Currently demand for it overwhelmed it’s supply, so it’s hard to come by, but I was lucky to get two development boards already.

So. What’s it good for? Well. There are many projects by hobbyists and geeks already in the workings, but since it runs on fully capable Linux it is very good solution for many things, especially universal and powerful robotics controllers. I have few ideas for projects using my Raspberry Pi’s. I want to talk a little bit about one of them here.

Since I have a recreational plot in the countryside, there’s always a problem with grass growing fast. On this parcel there are some flower borders, some bushes and some fruit trees. Also the terrain is a little bit rough. Mowing grass on this parcel is a lot of work and it needs to be done almost weekly, because grass grows really fast. Unfortunatelly neither I, nor my parents have time to do this. And since our lawn-mower is somehow old and mowing pieces of rough terrain requires taking mowed grass to the composter, it’s tedious and hard work. So. I have come up with the idea for automatic and robotic lawn mower. I have already started programming side of the project and am currently investigating mechanical and electronic solutions for this project. Since I own a car for few months now and had to work on fixing it I have acquired much knowledge about mechanics, because I am forced to repair and bring my car to usable state by my own – fairly saying because I don’t have money to let professional mechanics do it.

Let me talk a little bit about this project. I have already coded some basic building blocks like discrete topographic representation maps for the terrain. My language of choice is of course Python. The terrain map is a two-dimensional representation of discrete areas – one cell in an array represents ten by ten centimeters of terrain area, which should be quite sufficient, but resolution is adjustable and is limited only by amount of memory and computing resources needed for it. Since the topographic map is discrete representation of square 10 cm cubed areas and I need to represent only passability of the discrete area at hand, I have developed a class called BitMap which uses a great bitarray module. It’s lightning fast and uses very little resources. For example. A representation of 64 by 64 discrete cells which corresponds to 6,4 by 6,4 meters area takes only 512 bytes memory space. So basically representing a large terrain of thousand square meters with resolution of ten centimeters would take 12,5 megabytes of memory, which is not really a lot, given memory constraints of Raspberry Pi. Of course this topographic map can be divided into segments (regions) and offloaded to SD card and loaded on demand to conserve memory further.

The BitMap class is a represents passability of terrain chunks. Currently I’m developing fully software simulator for testing ideas that will be based on this class. The class allows loading and saving to byte representations, 1-bit bitmap images and implements few algorithms, for example Bresenham’s line drawing algorithm, ultra fast Queue-Linear flood fill, matrix combinations and differences. It also implements a simple bounding-box collision detection. I have few more ideas on improvement for this class, but for now it already fullfills it’s basic goals.

The idea for my robotic lawn mower is to put it in unknown terrain and allow it to map it and it’s boundaries. Mechanically robot will be equipped with a gasoline powered engine like in typical lawn mowers. The engine will power alternator that will feed current into battery. The battery will be used to power electronics (including Raspberry Pi controller) and used for electric starter motor for the gasoline engine. Engine will be connected to simple electronic driven clutch and two-gear gearbox (forward and backwards). Front wheels will be able to turn by a servo controlled with Raspberry Pi. There will be some ultrasound sensors mounted on the front and back of the robot to detect obstacles. Since the lawn mower will explore the terrain by itself it will have some mechanical or other type of sensors for detecting holes in the ground, so the robot won’t fall into them. I haven’t decided just yet what solution I will go with for this problem. There’s also a problem for detecting off-limits areas of the terrain like water reservoirs and flower areas. There will be a camera mounted on the servo to allow computer vision including, but not limited to shape detection, obstacle detection aid and entity detection.

Since the robot must also be careful not to harm any animal or human in the area of robot operation (we are dealing here with quickly rotating knifes) the camera and sensor arrays will also aid in detecting interested cats, playful dogs, humans, etc. This will need careful programming of threshold values. It also means that terrain mapping will need to utilize some kind of heuristic allowing for exploring chunks of terrain previously mapped as inaccessible. Since the problem with lawn mower is covering unkown terrain area of operation as quickly and efficiently as possible, topographic mapping and detecting area limits is of crucial importance. The software simulator I’m currently building will allow me to test different navigation and area covering algorithms, but also will provide me with the platform for implementing statistical and feedback based neural networks, allowing the robot to learn and improve decisions on operation algorithms with each iteration for given terrain. Pathfinding will be based on heuristic algorithms including graph based D-Star which is successfully used on Mars Exploration Rovers and military grade autonomous systems.

Since Raspberry Pi is equipped with sixteen GPIO ports and also I2C bus, designing a relay board for sensor arrays and servomotors shouldn’t be problematic. Of course connecting sensitive electronic board to electrical parts can be quite dangerous, filter systems must be also implemented to protect the controller board.

That’s all there is for now, I will post more details about this project will be available as it progresses. So stay tuned, comment and wait for further updates.

About Wolverine

If you are looking for IT consultant, let me know! karol at karoltomala dot REMOVE com Just remove the REMOVE word from the e-mail above!
This entry was posted in Hacking, Programming, Python. Bookmark the permalink.

2 Responses to Robotic Raspberry Pi powered lawn mower

  1. fredrik says:

    Hello intersting project, have you looked into what otherhardware to use? my self I was thinking of phidgets for sensors and movement

  2. Wolverine says:

    Well. This project is currently on hold, because as always with hobby things, time is lacking. I have thought about using arduino board for it, but I wanted to prototype with RPi since it gives you a broader flexibility. Never heard of phidgets, would certainly look into them.

Leave a Reply

Your email address will not be published. Required fields are marked *