Jean Luc Farrugia has researched an intelligent robot control system which autonomously avoids road hazards and the deadliest traffic accidents.
It is estimated that every year, 1.2 million people die and another 50 million are injured by traffic accidents (WHO, 2012). Driver error is the sole or a contributing factor in 60 per cent and 95 per cent of accidents respectively.
Computers are not prone to stress, fatigue, distractions... and can react faster with a broader environmental awareness- Jean Luc Farrugia
The situation doesn’t look like it will improve – according to a 2009 Yung-Hsiang report, such accidents are expected to increase by about 65 per cent over the next 20 years unless more accident-prevention research is carried out.
Of course, improved driving skills and better quality roads can all help to reduce such statistics. Yet technology can also play an important role in the prevention of traffic accidents. The reason is that technology is more efficient than human drivers – unlike human drivers, computers are not prone to stress, fatigue, distractions or other conditions, and can react faster with a broader environmental awareness.
In research that I conducted as part of my undergraduate degree at University, I built a robot that can be controlled from a remote computer over Bluetooth. I also investigated the design of an intelligent robotic control system which can autonomously avoid hazards, including collisions, and rollovers, as a proof of the concept.
Rollovers, considered among the deadliest of traffic accidents, are related to the stability of the vehicle in turns, which is determined by the relationship between the centre of gravity and the track width (distance between the left and right wheels).
A narrow track width and a high centre of gravity can make a vehicle unstable in fast turns or sharp changes of direction, making SUVs and trucks more vulnerable. Rolling initiates when the centripetal force to make a turn is enough to cause an upward force which is greater or equal to the vehicle’s weight. A vehicle will topple over when rolling persists until a vertical line drawn from the centre of gravity falls outside the vehicle base.
Traffic accidents can be deadly and also incur high damage costs – this can be seen in the terrible multiple-vehicle crash that happened in 2011 on the M5 motorway in the UK, and another 2011 crash involving 14 supercars in Japan.
Traffic accidents can be harmful to both the occupants in the vehicle concerned as well as to pedestrians or colliding objects. A collision and roll avoidance safety mechanism is by no means limited to smart vehicles, but is also viable for various transport and mobile robotic systems such as tele-robotics – the latter refers to a situation where an operator remotely controls a robot over a communication system, with limited situational awareness, smart wheelchairs, marine vessels, and more. People who might not have fine motor coordination or fast enough responses would especially benefit from such a system.
The need for hazard avoidance mechanisms in tele-operated robots is evident – this is because the operator not only must cope with challenges inherent to driving, but must also do so with a restricted field of view, limited depth perception, disorienting camera viewpoints and significant time delays, leading to high operation failure rates.
Moreover, human operators tend to oversteer and lose control when delays are introduced. Examples of accidents which occurred when tele-operating a robot include accidental piercing of a wall on a search and rescue mission and a robotic construction machine which accidentally hit an oxygen cylinder, causing the latter to explode.
Mobile robots have been designed for use in a wide range of sectors, including domestic, medical, military and scientific exploration. Tele-robotics echo artificial intelligence pioneer Marvin Minsky’s vision of a remote-controlled economy – ‘tele-presence’.
Forms of technology that warn drivers of impending dangers may not be ideal as many drivers panic and fail to act when a collision may happen – in some cases, drivers act in ways that do not help to lessen the chance of a crash.
Current braking and roll stability technologies, such as ABS, ESP and RSC are reactive – that is, they must detect a problem in order to undertake corrective action. On the other hand technologies that can anticipate and prevent loss of control can offer better safety than reactive approaches.
Reactive techniques are undesirable especially for rollover prevention – this is because stability rapidly gets much worse when rollover begins and recovery is very difficult even for professional drivers. In addition, current roll stability control systems are unable to prevent rollovers caused by sudden turns at high speed.
On the other hand a fully autonomous driving system is especially prone to the ironies and paradoxes of automation, whereby the more automated a system is, the more critical is the user input when the system fails and yields control to the operator. On the other hand, the more reliable and automated the system is, the less active is the participation of the user in the running of the system.
Hence it is then more difficult for the operator to remain attentive and perform correct actions when the automation system fails, as they would have become used to the system performing well, where their role is downgraded to a mere observer. Consequently the operator becomes detached from the system and the skills needed to control the latter deteriorate as they are not put into practice.
“The really hard things to automate or synthesise, we leave to the operator to do,” says Ericka Rovira, an assistant professor of engineering psychology at the US Military Academy at West Point. “That means people have to be alert and ready to act at the most crucial moments, even though the monotony of monitoring supposedly reliable systems can leave them figuratively or physically asleep at the wheel.”
TCAS (Traffic Alert and Collision Avoidance System) and Envelope Protection (also known as carefree handling) approaches used in aviation are also of interest as the pilot still retains involvement – moreover, pilots are free to avoid collisions in advance, or fly the aircraft as desired, as long as their inputs conform to the controller set limits, preventing the aircraft from entering a dangerous state such as stalling.
In my research, collision avoidance was achieved by constraining user control inputs within a safe dynamic envelope which continuously changes according to the ultrasonic sensor readings.
Hence the driver or operator is allowed to freely drive the robot when obstacles are not in the vicinity, but control limits are enforced as obstacles are encountered or approached, constraining driver input which would have resulted in a collision, and allowing input which abides by the current limits.
Rollovers are avoided by limiting the speeds at which turns are made, preventing the vehicle from exiting a safe driving envelope, which would otherwise have resulted in toppling over. If necessary, vehicle speed is autonomously decreased before turning and hence rolling is prevented before it has even started.
The system can autonomously control and limit acceleration and deceleration, throttle and steering. Motor settings can be based on purely user values, machine values or a combination of both, according to different circumstances.
When there is not enough space to steer away from obstacles, the system chooses to reverse, and maneuvers similar to the three-point turn are executed, providing assistance in navigating out of tight and restrictive areas.
Pipelining and parallel programming were used to enhance response times which deteriorated with the presence of communication latencies. To evaluate the collision avoidance system, external testers were instructed to deliberately try to crash the robot in different and unseen environments. The system seamlessly constrained or took over user input in collision-prone situations, while allowing the user to drive freely in open plains.
Rollover protection successfully limited the throttle when steering, which resulted in the robot toppling over when it was deactivated.
The system provided a robust performance in the face of variable communication latencies, inaccurate sensing, and hazardous user-input, adapting to unseen and varying environments and situations.
The safety advantages of an automated driving system such as the Google self-driving car are employed while driver involvement is retained, thus avoiding the harsh consequences of the automation paradox and giving the driver freedom of control as long their actions do not pose any danger.
Jean Luc Farrugia is a B.Sc (Hons) ICT (CSAI with CCE) graduate from the University of Malta. His thesis project, entitled ‘Hazard Avoidance Auto Control System for a Robotic Vehicle’ originated as his idea. The thesis was supervised by Dr Matthew Montebello and Dr John Abela from the Faculty of ICT.
Road safety
The European Transport Safety Council’s annual PIN report says that Malta lags behind when it comes to planning, managing and implementing road safety strategies.
The report notes how Malta is one of eight surveyed countries without a national road safety plan, and says Malta has no budget dedicated to the implementation of a national road safety programme.
However, despite the negative findings, statistically Malta’s roads remain among the EU’s safest, with road fatality rates below the EU average.
The ETSC is an independent non-profit making organisation based in Brussels – it is dedicated to reducing the number of transport-related deaths and injuries in Europe.