Autonomous Ground Vehicle Mk IV

Projects‎ > ‎

Autonomous Ground Vehicle Mk IV

Autonomous Ground Vehicle

for weekly progress please see: Weekly Progress

YouTube Video

First Outdoor Test

Project Background:

Robotics is the branch of technology that deals with the design, construction, operation, and application of autonomous machines that can sense and modify their environment. It is a rapidly growing field of both professional and amateur researchers and developers. However with the exception of a select few, practical robot platforms and sensors for an outdoor environment don't exist. While there has been developments focusing on robot applications in the outdoors, the major of these have been focused on big budget robotic applications on very large platforms (automobiles, etc.) or military applications due to the cost and challenges inherit in the outdoor environment. Small robotic platforms lack the power or clearance to move over high friction rough terrain, and the environment itself is a hazard to the electronics and mechanical systems. Dust, water, mud, gravel, and sudden drops are all common features of the outdoor environment which can damage a robotic platform. Pedestrians, cars, animals all present new and dynamically changing obstacles which are difficult to detect and avoid. The goal of this project to develop a chassis that can overcome the terrain obstacles and be modular enough to easily incorporate improvements to provide a platform on which the sensing aspects of outdoor robotics can be developed and tested on. Currently no such platform under $10,000 exist.

Educational Value Added:

-This project supports further interest in robotics, especially UGV and other specialized, outdoor robotic systems.

- Experience is directly applicable to classes, dealing with microcontroller interfacing, communication protocols, statics, dynamics, and power systems.

-Novel concept and implementation of a fully articulated suspension system, allows high speed motion over difficult terrain

Members:

William Gerhard

Operational Requirements:

Specifications
Ground Clearance	> 3 inches
Ground Pressure	< 1.5 PSI
Obstacle Crossing	6 inches
Vertical Trench	5 inches
Speed Cross Country	4 mph
Maximum Incline	50 degrees
Battery life	> 1 hour (typical use)
Carrying Capacity	> 10 pounds
Weight (Empty)	< 90 pounds

Current Status:

Was functional, current not work. Cause unknown, PWM from the PRU is not being transmitted to the motor controls. However when the Beaglebone Black is removed from the robot and placed on a desk, PWM works. Also when the motors start, the Beaglebone resets. The system was functional during June 2015. Since the Beaglebone Black cannot run the monocular camera obstacle, so the embedded computer needs to be replace regardless. And a new battery is required, since the current battery was damaged in shipping.

Design Concept:

Version 1.1

Version 1.0

Implementation:

Version 1.0

Material Selection:

See :Material Selection

Caterpillar Treads:

Caterpillar treads have two major advantages over wheels they greatly reduced ground pressure and a much larger ground contact area. The Dagu Wild Thumper is a 6 wheel drive supposedly "all terrain robot" of a much smaller size than this project. It has approximately 11.5 square inches of ground contact, and with a maximum weight of 11 lbs, its ground pressure is 1 psi. The AGV has approximately 112.5 square inches of ground contact and a maximum empty weight of 90 lbs, exerts a ground pressure of .8 psi. The larger ground contact means a larger area to distributes the force from the motors, reducing the chance of bogging down is soft terrain.

However the treads do have disadvantages, the larger surface area means more static and rolling friction, requiring more powerful motors. The tracks also have the potential to come off, immobilizing the vehicle.

Suspension system:

Having a functional suspension system is a mission critical component. Without a suspension system, the AGV will never be able to reach its maximum top speed in real world conditions and the vibration from traveling at any speed over difficult terrain will reduce the lifetime of electrical components. See video below for the difference a suspension system will make when crossing rough terrain at high speed.

Locomotion Difference

Mechanics:

Suspension system:

Revision 1:

exploded view of single suspension module

Suspension System Model

This above video demonstrates the mechanics of the suspension system. Note, there is a spring connection between one of the bolts on the central hub and the bolt at the top. The advantage of this system over a traditional torsion bar or single pivot design is threefold. First, the main vehicle hull is not not compromised. This means a reduced chance of foreign material or water affecting the electronic systems inside. Second, because there are five bolts attaching the wheel supports to the rest of the suspension system, the chances of a side impact damaging the suspension system is reduced. Third, the entire system is self contain which makes repairs, replacements, or upgrades as easy as removing five bolts. The only disadvantage is that mud or rocks could potentially get lodged inside the slots and jam the module. Testing will see if this is in fact an issue.

Problems:

Failure to take into account the forward center of gravity and under-estimation of the final system weight meant the suspension system was too soft, resulting in it's inability to support the AGV mass and prevents the suspension system from returning to an "ideal" position.

Revision 2:

Prototype made from plastic to test tolerances and fit

Revision 2 of the suspension system address several issues which hampered the effectiveness of the first suspension system. The re-curved design reduces the angle between the pivot point and the ground contact, reducing the load on the spring when in equilibrium. The new design also raises the wheel by .724 inches, reducing the amount of torque the spring needs to return the wheel to the upright position. In addition to mechanical design changes, the new suspension system pieces will be made of 3/16" aluminium rather than steel, which will reduce the overall system weight. Theses features combined should resolve all issues with the suspension system. This system was successful, see video at top of page.

Revision 2.5:

Revision 2 of suspension system worked, however in certain cases the track would pop off of the wheels in the suspension system and become stuck. The best solution to this problem was to add an addition plastic piece outside of the wheel which would help keep the track in-line and on the wheels. These parts were 3D printed out of PETG and bolted on. They have significantly reduced the number of thrown tracks.

Hull:

The AGV hull is composed of aluminum panels that are bolted to a frame made from 80/20 extruded beams. This creates a strong, lightweight and modular hull which can easily be repaired or replaced. Every element, even the suspension system, bolts into the 80/20 frame.

Center of Gravity:

Tracks:

Old:

The tracks will be composed of Intralox series 900 sprocket driven conveyor belt. These are an industrial grade plastic chain composed of individual plates, which are riveted together. Overall, they appear to have good abrasion and wear resistance, and a large surface area to reduce ground pressure.

image from Intralox

New System:

In the continual goal to improve performance and increase the system speed and ability to cover terrain, the tread system needs to be improved. The current conveyor belts work, however it had inadequate performance on smooth floors, going up curbs or cliffs, or going through mud and rocks. Attempts were made to modify the existing system, but Delrin is designed to be smooth and non-stick, and it didn't work. In addition, the conveyor belt cog meshes well, until it slips and pop's out of the texture in belt or bind in the texture, which is not helpful.

By utilizing a standard ANSI-35 chain as the base, there is a much wider range of options for keeping the chain captive and driving the chain.

Drive Train:

Previous Implementation:

The drive train will consist of two Bag motors commonly used in FRC robotics competitions connected to VersaPlanetary gearboxes, which allows a variety of gear ratios to pick the optimal configuration for the intended application. The original design utilized two Bag motors, spec shown below.

BAG Motor Specs

Free Speed: 14,000 rpm (+/- 10%)

Free Current: 1.8A

Maximum Power: 147 W

Stall Torque: 3.5 in-lbs [0.4 N-m]

Stall Current: 41A

Current Implementation:

The dual input gearbox and increasing weight of the system made these BAG motors impractical. As a result, the more power Mini-CIM motors were selected to replace them.

Motor Spec:

Free Speed: 6,200 rpm (+/- 10%)

Free Current: 1.5A

Maximum Power: 230 W

Stall Torque: 12.4 in-lbs [1.4 N-m]

Stall Current: 86A

With a 1:28 gear ratio this should give the AGV approximately 350 lb of torque combined on 4" sprockets, with an rotating speed of approximately 221 RPM and a top speed of 4 mph cross country.

The change in motors does reduce the top speed, but triples the amount of torque available. For the time being this is a reasonable compromise, but different gearboxes/motor combinations could be considered to improve performance.

Battery:

Current testing shows a 12 AH battery provides approximately 25 minutes of run-time with BAG motors. This projects to at minimum a 40 AH battery to run the system for the required 1 hour. Lead acid chemistry appears to be the only choice that can provide the required surge current (100+ amps).

Selection process is on going.

Speed Controllers:

Old:

The Jaguar Motor Controller provides variable speed control for both 12 V and 24 V brushed DC motors at up to 40 A continuous current. Includes a wide variety of sensor interfaces, including analog and quadrature encoder interfaces, high performance Controller Area Network (CAN) interfaces, and an RS232 port as well as PWM control.

With the use of much larger motors, these speed controller may no longer be a viable control selection. However, with the built-in over current protection these should be adequate for testing purposes.

New:

Victor SP

The Victor SP has been redesigned from the ground up, and the latest iteration features completely overhauled internals and an innovative case less than half the size of previous models.. Features include:

Full aluminum case, passive cooling fins, and sophisticated internal components make fans optional
Completely sealed enclosure prevents debris from getting anywhere it shouldn't be
Electrically insulated components allow for direct controller mounting to a robot frame with no fear of shorting
LED indicators blink proportionately to output speed for easier debugging
Illuminated "Brake / Coast Calibration" button enables one-touch setting changes and calibration
Robust embedded power & output cables will never shake loose during a match
Built-in pockets sized for #8 nuts and zip tie grooves make rigid mounting simpler than ever
Integrated sign-magnitude synchronous rectification reduces heat generation at full motor stall
15 KHz output switching frequency ensures smooth and precise motor control

This new controller will allow the motor controller to be hard-mounted to the frame and used the aluminium skin of the robot as a heat sink.

Electronics:

Original Proposed Architecture:

Control board will be a Beaglebone Black running Debian. The Beaglebone black will handle all motion control, sensor reading, and communication to the external world. Some small subsystems will be independent with and contain micro-controllers in order to interface with systems that may require real-time control or non supported communication interfaces.

Communications:
The Beaglebone Black will support two physical layers. 2.4 GHz IEEE 802.11 and 900 MHz.
Wifi communication will provide short range communication and real-time debugging and the longer range 900 MHz will allow some limited form of real-time data and other information event at long range. Remote operation commands can also be sent using the 900 MHz channel.

Problems with the Beaglebone Black:

Limited Computational Power:

While the Beaglebone black has a very nice SOC and hardware design, its single core 1 GHz processor cannot run complex image manipulation and object detection algorithms fast enough. The fundamental limit discovered while working for McQ. was 5 frames per second. In order to reach that speed, the code had to high optimized and and complied with all hardware optimization options. It was difficult to do and is not conducive to rapid algorithm development, nor do I think 5 FPS is sufficient to ensure the robot does not hit an obstacle.

USB is broken:

With the 3.8.XX Kernel of Debian the Beaglebone comes with, USB is not fully implemented. It does not support hot swapping and I do believe there are issues with transferring images over USB, (It should not take ~1 second to receive a 320 by 240 photo using the V4L API). In addition, the removal of USB devices is not detected and any subsequent attempts to read/write to that device will cause the system to crash.

Difficult to understand kernel changes:

While the kernel is actively being worked on, there are several different repositories and no change logs on the newest repository, so it is difficult to know what features work or do not work.

Unimplemented features/Broken Features:

Some of the nice features in the AM335x processor are not implemented or do not work. CAN bus does not work in the 3.8 kernel (at least that I have been able to find), The quadrature encoder hardware does not work, encryption acceleration hardware does not work, the PRU units are vaguely working in the standard distribution (with kernel patches and some work), automatically load device trees for custom capes does not work (works for off the shelf capes already complied into the kernel), and the overall result is all of the features that would make this a nice platform for a robot are not working.

Overall conclusion:

Since in the end I am a electrical engineering and not fluid in linux by any stretch, it is beyond the scope of this project to get the features I want to work properly on the Beaglebone black. The more I played with several single board computers, to more I realized that a singleboard computer is not a really powerful micro-controller. In the end the role should be computational intensive tasks and managing smaller sub units which do specific tasks. For example instead of generating a PWM signal on the single board computer, a standard I2C pwm extender can be used, which is a lot easier and gives a cleaner PWM signal.

New Architecture:

Computer Selection

PCB Design (Obsolete):

PCB Design

Sensors:

For most recent sensor information and design, see Sensors

(Obsolete)

Obstacle Detection:

Long Distance:

LIDAR Lite

Specs:

Range: 0-40m Laser Emitter

Accuracy: +/- 0.025m

Power: 4.7 - 5.5V DC Nominal, Maximum 6V DC

Current Consumption: <100mA continuous operation

Acquisition Time: < 0.02 sec

Rep Rate: 1-100Hz

Interface: I2C or PWM

This unit will give the ground vehicle a long range sensor able to detect obstacles far head reliability in an outdoor environment. The unit is a class I laser unit, and is therefore usable around humans and animals with no risk of eye damage.

Medium Range Detection:

To Be Determined, high probability of some form of ultrasonic sensor array for coarse obstacle detection and limited avoidance of rapidly changing obstacles, I.E. people, dogs, etc.

Visual Detection:

C920 Logic Tech Video Camera

Supports H.264 video encoding which reduces CPU load on BeagleBone Black
Full HD (1920 x 1080)

Color Detection and Tracking:

A CMU Cam V5 will provide limited object recognition and tracking for specific objects and allow the ability to determine contextual clues through color codes, I.E. able to see where the charging station is. Advantage of this system is all processing is done on the board, minimizing load on the Beaglebone Black.

Another option is to experiment with OpenCV pedestrian detection algorithms running through a HD USB Camera. However running a higher level algorithm may require more a power control platform.

Spacial Orientation:

System will include a full 9 degree of freedom IMU, which includes magnetometer, accelerometers, and gyroscopes located at the platform's center of gravity to provide accurate information about the systems spacial orientation.

Location:

Platform will include one GPS to allow the AGV to determine its location within a realistic 5 meters. Combining rough position and some knowledge of topography and long range sensor data could help plan routes around obstructions.

Other:

TBD

Ruggedizing:

In order for the platform to truely function outdoors, water and dust must not effect the electrical systems inside. Instead of trying to seal the entire chassis, which would be difficult and make replacing or modify the frame more challenging, the current idea is to use IP66 or high boxes and connectors inside the frame to house the motor controllers, beaglebone black, supporting electronics, and sensors. Theses boxes keep everything safe from dust, impact, and water, and cleans up the internal wiring.

So far the only downside is cost, but compared to alternatives it seems to be the ideal option.

Requirements:

Jaguar Motor Controller is 3.5 inches by 4.5 inches by 3 inches tall.

1555FF42GY

ABS IP67 enclosure. 4.722" L x 3.574" W. Should snugly hold the motor controllers and give plenty of room for the beaglebone and supporting electronics. (Separate box for each system).

Large enough to hold PCB from the C920 Web camera. Enclosure is IP66. Transparent ABS allows for enclosed optics.

Here is the camera mounted in the IP66 enclosure:

Connector:

HR30 connectors. Cheapest waterproof connectors, 12 pins gives widest range of IO possibilities. Can run PWM, low current power, I2C, CAN, and UART. The connectors are IP67.