Fullscreen HTML5Med Res (??MB)HTML4

University students in Australian competition . . . Autonomous Vehicles by Dr David Maddison Over several days late last year the inaugural annual Autonomous Ground Vehicle Competition (AGVC) was held near Geelong. Open to tertiary students from throughout Australia, it was the first such event to have been held in Australia and its underlying theme was “Autonomous Navigation”. T he event was held at the Waurn Ponds campus of Deakin University and was hosted within the Centre for Intelligent Systems Research (CISR) (see www.deakin.edu. au/research/cisr/index.php). The primary sponsor of the event was Australia’s Defence Science and Technology Organisation (DSTO) whose purpose was to promote technological development in the field of robotics in areas in which there were perceived deficiencies in Australia (see box). One of DSTO’s interest in robotics is to improve the effectiveness and safety of Australian soldiers by having semiautonomous robots relieving them of “dirty, difficult and dangerous” tasks. Examples of such tasks are defusing of improvised explosive devices as used in Afghanistan, going into contaminated environments, carrying heavy loads, gathering intelligence such as reconnaissance and surveil14  Silicon Chip lance; and detecting, designating and even destroying enemy targets. Currently, much military robotic technology requires “tele-operation” with operators having complete or almost complete control of the robot. While it is still considered desirable for a human operator to have ultimate Indicative map of Qualifying Navigation Course. (Based on US IGVC [Intelligent Ground Vehicle Competition] course.) siliconchip.com.au command of the machine and to make critical decisions such as when and where to engage an enemy target, there is great scope to make a robot more autonomous in many of its activities. For example, if a robot was required to navigate to a certain location, rather than a human operator guiding every turn of the vehicle, it would be more desirable for the operator to simply instruct the vehicle as to the final destination and the robot would decide the appropriate route to take. This would result in a change of the human operator being (to use DSTO’s terminology) “in-the-loop” to them being “on-the-loop” with ultimate command but the operator being relieved of small decisions and a significant workload. Such higher levels of autonomy cannot be achieved without more sophisticated algorithms for sensing and decision making. The DSTO’s sponsorship of the AGVC aims to encourage the development of such technologies in Australia. Competition The AGVC consisted of three components: Technical Qualification of the robot; the Autonomous Navigation test; and the judges’ evaluation of the robot design (Design Competition). Why hold the AGVC? The event was intended to explore and develop technologies that will result in improvements in autonomous vehicle related areas, within Australia in which there is a current perception of a deficiency in the following areas: • • • • • • • • • • • • • • • • • • • • Sensor data fusion Image and sensor data processing Target recognition Artificial intelligence Knowledge based systems Open system architectures Machine vision Autonomous navigation and mapping Modelling and simulation Human-machine interfaces and integration Computer hardware and software designs Mechanical and electronic architectures and systems Communication networks Developing fast search algorithms Multi-vehicle coordination teaming algorithms Hardware sensor systems Real-time computer hardware and software systems Higher level of autonomy Multi-robot collaboration Target identification and classification. Technical Qualification was designed to ensure that the robot met certain technical and safety standards and could navigate a qualification course. Such standards allowed for: • either a commercial robot chassis or a custom designed one; • that the vehicle be a land-based vehicle with either wheels, tracks or be a hovercraft etc; • that it fall within a certain size range; it be electrically powered; A section of Qualifying Navigation Course showing grassy surface (significant, because the uneven texture of the surface causes greater difficulty in implementing edge detection algorithms, especially as the sun angle and cloud cover change), the white lines denoting the sides of course lanes and different coloured barrels as obstacles. siliconchip.com.au April 2014  15 Indicative map of Autonomous Navigation Test course. Note the GPS waypoints set among numerous obstacles in the centre section. • it be hardware controlled (not software controlled); • have a mechanical stop button and also a wireless operated stop button (also not software controlled) for safety reasons; • that it displays a safety light to indicate that the vehicle was both powered and in autonomous mode; • it had to be able to carry a provided 9kg payload. Apart from Technical Qualification, the robots had to complete a Qualify- ing Navigation Course to be accepted into the final competition (Autonomous Navigation Test). The Qualifying Course was laid out within an approximately 30 x 60 metre grass area and included a track comprised of a pair of painted lines containing straight lines, curves and barrel-shaped obstacles. Tests that had to be passed were to meet a certain minimum speed requirement of 1.6km/h (maximum speed 16km/h) and to demonstrate lane following by tracking between the marked white lines, obstacle avoidance and the ability to meet a GPS waypoint by navigating around an obstacle. If all tests were passed the teams could progress to the Autonomous Navigation Test. Additional rules included: • that vehicles must be unmanned and autonomous and must compete based upon their ability to perceive the environment and avoid obstacles. • that they cannot be remotely oper- At left is an example of obstacle navigation, where the vehicle must negotiate all of the barrels without hitting them, at the highest speed it can manage. At right is a similar shot, this time navigating to GPS waypoints. 16  Silicon Chip siliconchip.com.au A variety of shapes and sizes of Autonomous Vehicles was designed and built by various Australian university students. There was even one based on an electric Personal Mobility Vehicle (overleaf)! ated by a person during the tests all computation, sensing and control equipment must be located on the vehicle and no base stations to improve positional accuracy were permitted (although the use of differential GPS [DGPS] was allowed). No “remote control” Vehicles were allowed to be remotely operated so they didn’t have to be carried to the start line but that remote operation mode had to be confirmed to be disabled before the start of the competition. The Autonomous Navigation Test siliconchip.com.au was somewhat similar to the Qualifying test but with more complexity and a greater number of rules. There was still a minimum speed requirement, many more obstacles, the lane edges could be marked either as continuous lines or dashes, the track width was variable from three to six metres wide, there were inclines and flags to navigate between in the latter part of the course, eight GPS waypoints to navigate to and an increasing level of difficulty as the course progressed. The first third of the Autonomous Navigation Test comprised of two white lines forming a track on the grass field, various obstacles (ingeniously and inexpensively made from painted compost bins!). After the first GPS waypoint there was “No Man’s Land” where there were no structured lines and there were fences and obstacles that the robots had to navigate around to get to an additional seven GPS waypoints. Within the No Man’s Land there was a “Money Barrel”, the locating of which would entitle a team a trip to to the IGVC in the USA. If the robots made it to all eight waypoints they could then enter the final third of the course where they encountered April 2014  17 LEVELS OF AUTONOMY There are no strict definitions of what is meant by robot autonomy (or robots for that matter) but three basic working definitions of autonomy might be considered. 1) Tele-operation. A robot responds only to direct human command. A radiocontrolled model car or robot to defuse explosive devices is an example. 2) Semi-autonomous. A robot is controlled by a human but can perform basic tasks. Automatic parking or automatic braking upon imminent collision in some cars are examples. 3) Fully autonomous. A robot is given a task to perform and it does so until countermanded by a human. The CIWS weapon system mentioned in this story is an example. Of course, any given robot could be operated in any of these modes as required if it has the capability. another set of marked white lines and further obstacles and flags they had to go between. Various penalties could be issued by the judges such as for holding up traffic, leaving the course, vehicle crash or obstacle displacement, careless driving, side swipe or obstacle touch, student’s choice to electronically stop, judge’s choice to electronically stop, blocking traffic, loss of payload, passing on the wrong side of a flag and running over a flag. The most severe penalty was for going too slowly which resulted in disqualification. Prizes Ten teams arrived for the competition from all over Australia except Tasmania and the Northern Territory and of these, eight qualified to move onto the final stage. Robot Operating System (ROS) is a Unix-like software framework design for robotics. It was first developed by Stanford Artificial Intelligence Laboratory in 2007 and remains under development by many groups as an open source software project released under a BSD license. ROS has two basic parts, one part is the operating system which provides traditional operating system services and the other is a collection of user-contributed packages that provide functionality specific to the research interests of the research group that provided them. For example, a research group might specialise in mapping, 18  Silicon Chip Judges ranked the entries according to a combination of time taken and distance progressed if the vehicle could not complete the course. Trophies and cash prizes were available to the top three teams if they completed or substantially completed the Autonomous Navigation course as follows: First Place: $15,000 plus two economy return airfares to compete at Intelligent Ground Vehicle Competition (IGVC) in the USA (www.igvc.org/) and $2,000 towards expenses, Second Place: $10,000 Third Place: $5,000. If no teams completed the course the prizes were first, $5,000, second $3,000 and third $2,000. Unfortunately, despite an outstanding effort by all teams, no one managed to complete or substantially complete the course another group may specialise in planning or machine vision. This collaborative approach to software development is essential because problems that seem trivial for a human such as picking up and cracking an egg into a fry pan can be enormously complex to implement in software and no single research group can hope to write software that masters all tasks. The ROS collaborative model allows groups to share and build on each other’s work and this allows more efficient software development. ROS has an architecture based on nodes which individually receive and process sensor, actuator and other data and which can communicate with each other. The ROS library supports Ubuntu Linux while there is experimental support for Fedora and Mac OS X. As well as the experimental robots, ROS has been incorporated into many commercial models. siliconchip.com.au this year so the lesser amount of prize monies were awarded. It should also be noted that extremely minor issues could constitute the difference between success and failure so a technical “failure” should by no means be considered to reflect poorly on any team. Finally, there was the Design Competition stage in which judges examined design innovations within the vehicles. This required a full written report by the students as well as a presentation before an expert panel. Prize money for this was, at the discretion of the judges, up to $5,000. Also, at the discretion of Organising Committee, research grants of up to $1,500 were available for any team thought to have a particularly interesting approach. Teams and their vehicles The teams were highly dedicated and had typically taken six months preparing the vehicles, all on their own time and often with their own money and all in addition to the high workload of their university studies. Furthermore, no course credit was received by the students for their work but hopefully that will change in future years. Dr James Mullins, one of the organisers from Deakin University said “People are very passionate about what they’re doing. It’s really great to see this happen in Australia. We’re seeing a lot of people that previously wouldn’t have had as much applied knowledge from research fields being able to put their technology into diverse fields such as machine vision, vision sensors, algorithms, inertial measurement, GPS, and even the base mechanical platforms.” As the team are undergraduate-based with relatively little team sponsorship in this first competition they had to struggle with low cost technologies. Others chassis included the commercially-available Husky A200 from Clearpath Robotics. Another interesting chassis was based on a personal mobility vehicle (see above). Computing systems ranged from notebook computers installed in the robots, pc-type computers intended for installation in cars through to embedded logic such as Field Programmable Gate Arrays (FPGAs). Operating systems used on the robots included various versions of Linux with Robot Operating System and Microsoft Windows, Most teams used Vision Systems cameras for line edge detection with OpenCV as the software library and most teams also used Lidar for obstacle detection in which a laser scans the environment to build a three dimensional model of the surrounds. (Incidentally, Lidar is not an acronym The vehicles The vehicles consist of several main parts: the chassis, the computing system, the software system and the the sensor suite. A variety of chassis were used. Some were free as donations (or low cost) such as the base component of an old electric wheelchair containing the drive gear and batteries, with some significant interfacing required to make it work. siliconchip.com.au Sadly, one robot crashed during the competition. At least the bins (normally used as course obstacles) were readily available . . . April 2014  19 Real Military Robots – Autonomous Ground Vehicles Military robots are in use right now by many military forces around the world. Most such vehicles are used in aerial operations and can, if necessary, operate with limited autonomy. Examples include unarmed Remotely Piloted Aircraft (RPA’s) such as the Israel Aerospace Industries Heron, as was used by Australian Forces in Afghanistan and armed Unmanned Combat Aerial Vehicles (UCAVs) such as the General Atomics MQ-9 Reaper or MQ-1 Predator as used by the United States in various theatres as well as many other Unmanned Aerial Vehicles (UAVs) in a great variety of forms with many different functions and capabilities. (See SILICON CHIP article “The Avalon 2013 Airshow”, May 2013.) In comparison to air vehicles there are far fewer types of Autonomous Ground Vehicles, possibly because navigation on the ground is far more complex than in the air, simply because there are far more objects and varying conditions on the ground that need to be taken into account. In comparison, most flying is done in a straight line and objects are relatively few and easily seen via well-developed sensors such as radar. A principle of unmanned armed vehicles and platforms is that there is always a human with ultimate command authority in charge. This remains true although the concept is a little stretched in the case of the Phalanx Gatling Gun Close In Weapons System (CIWS, pronounced sea-whiz). This is a relatively old US-developed weapons system (introduced 1978, but upgraded many times) that has been in use by Western navies, including Australia’s, for decades. It is intended as a lastditch defence when an enemy has penetrated outer layers of security. Once armed, it is programmed to automatically destroy incoming missiles, aircraft and other hostile incoming projectiles by firing 20mm rounds through a six-barrel gun at the rate of 4,500 rounds per minute with a muzzle velocity of 1,100 m/s. By necessity, once given authority to fire it must operate autonomously as with incoming supersonic projectiles at close range, there is no time for a human to react. It has been considered by the United Nations, unfairly, as a potential “lethal autonomous robot”. Apart from air and ship-based robots there are several land-based military robots in use. Early examples of unmanned tele-operated land vehicles included the Soviet Teletank and the German Goliath tracked mine, both of WWII although neither were considered great successes (see Wikipedia articles). Example of modern unmanned autonomous vehicles under development include the Lockheed Martin SMSS (Squad Mission Support System with “supervised autonomy” and used experimentally in Afghanistan), the UK/Australian BAE Systems MOATV (MultiOperated All-Terrain Vehicle) and the Boston Dynamics AlphaDog, a four-legged vehicle (search YouTube for “DARPA LS3” to see various videos) . These systems are intended in a troop support role such as carrying loads or medical evacuation of personnel. An Israeli company called G-NIUS has developed a series of security, patrol, combat support and combat AGV’s known as Guardium for a series of wheeled vehicles and Avantguard for a series of tracked Unmanned Ground Combat Vehicles (UGCVs). The vehicles feature autonomous mission execution, real-time obstacle detection and avoidance, fail-safe systems and off-road maneuverability. They can also carry extensive sensors suites as well and some can carry up to 300kg of soldier’s equipment or other payloads. Unarmed Guardium vehicles are in everyday use for border patrols and other activities. The Israeli G-NIUS Guardium Mk II AGV. Note the variety of sensors. 20  Silicon Chip siliconchip.com.au OpenCV (Open Source Computer Vision Library) is another open-source project like ROS and it can be used under a BSD license. It is a programming library for machine vision applications written in C++ and supporting many operating systems. as commonly thought but a combination of “light” and “radar”). Challenges The competition bought together students with a wide variety of specialised interests, encouraged teamwork and innovation and helped further establish a technological basis for autonomous ground vehicle development in Australia. There were many challenges to be overcome. One of the main challenges was detecting the white lines under a variety of lighting conditions, such as sun angles and sunny or overcast conditions as well as a varying texture of the grass or dirt surface upon which the white line was painted. Dr James Mullins said: “Challenges are in their vision systems. A lot of teams are trying to do this with relatively low cost technologies because they are undergraduates, it is the first year, they don’t have phenomenal amounts of sponsorship as yet ... so certainly the changing light conditions we’ve had over the last few days have been tricky but that’s why we’re looking for robust algorithms that can deal with that.” Despite the difficulty in line detection, obstacle detection was possibly more challenging so most teams spent most of their time on that area. An unwanted consequence of avoiding an obstacle was that the robot might go over a line, proving the importance of the line detection and the obstacle avoidance algorithms working in a cooperative manner. This year all of the vehicles worked in a “reactive” manner meaning they would respond to their immediate environment but had no knowledge of where they had been. So occasionally, a robot would reverse and return to where it had just come from or do a u-turn and get lost. Some teams were in the process of implementing mapping to overcome these difficulties but this feature was not fully implemented on any of the robots. siliconchip.com.au Originally developed by Intel in 1999, it is now is run by a non-profit foundation. It has an extremely strong user base around the world. OpenCV runs on popular operating systems such as Windows, Android, Maemo, FreeBSD, OpenBSD, iOS, BlackBerry, Linux and OS X. To find the eight GPS waypoints the teams needed a one metre GPS accuracy and most teams used highaccuracy differential GPS (DGPS) with a 70cm accuracy. The future It is planned that the AGVC will be run again next year, bigger and better. The teams were very excited about next year’s competition, already discussing what they want to do. Some teams are talking about fully implementing mapping so the vehicles have knowledge of where they have already been. It is expected that more teams will compete and that the level of technology will be greater. No doubt there will SC be some surprises as well! Winners are grinners . . . Winners of the Autonomous Navigation Test component, Order 66 from Deakin University. Team Name Affiliation Trial & Error ANU Order 66 Deakin Uni Dynamic RMIT Team 1 Team Redback Flinders Uni Aperire Incognitam Deakin Uni Team UQ Uni of Qld dUNSWiftly UNSW UWA Robotics Uni of WA Team Zelos Uni of Adelaide Team Tesla RMIT Team 2 Design Test Score 750.9 864.5 647.1 824.6 813.9 665.5 762 661.7 747.8 651.4 Auto-Nav Score Did not qualify 84 55.5 11 48 Did not qualify 108.5 21 5 39 Overall Ranking 6 1 5 4 3 0 2 0 0 6 Innovation Award: Trial and Error (ANU) April 2014  21