An algorithm for driverless cars in India

Driverless cars, then & now…

In 1999, when I had the opportunity to work on the design of an Autonomous Vehicle at Wayne State University in Detroit, our effort was primarily directed at ensuring the vehicle was keeping lane. So much research has happened since. Now, Tesla is claiming to release the world’s first production ready driver less car, well ahead of Google’s release.

Challenges for driverless cars…

Chaotic Traffic In India. Photo credit: http://en.wikipedia.org/wiki/Traffic_congestion

I’ve always wondered how this technology would work in chaotic driving and road conditions like India. I recently tried Assisted Parking on the Volvo v40 and must say it was pathetic. When lanes were clearly marked and ideal conditions simulated, It worked well. Else, it just failed to parallel park. Regards the Cruise Control, in all these years of driving, very rarely have I been able to drive more than a kilometer or two without breaking the cruise control for a strolling cow, dog or a passers-by that caught me by surprise. Suffice to say that even in case of Driver Assistance Systems like Cruise Control and Assisted Parking, substantial re-engineering is required to adapt any of these systems for places like India.

Adaptation of algorithms for driverless cars in India…

With Driverless cars, its a totally different ballgame. What algorithmic enhancements are required to the autonomous vehicle software systems to accommodate driving in an environment where there is absolute lack of standardization in road infrastructure, poorly planned road network, lack of lanes and road barriers, lack of on-ramps and exit systems, lack of directional, informational and warning signage, abundance of potholes and random speed breakers, poor mapping of roads and locations of interest and free-for-all, democratic access to roads for motorized and non-motorized vehicles, animals and pedestrians?

At the core of any driverless system is the ability to accelerate, brake and steer automatically based on inputs from the environment. At a very high level, the analysis and execution of driverless vehicular movement is coordinated by a Human Interface, Route Planning Module, Environment Perception and Modeling Module, and a Command and Control Module. At a more detailed level are capabilities for Forward collision warning, Lane departure warning and lane keeping, Pedestrian and obstacle detection, Traffic sign recognition, Headlight control, Terrain mapping, Blind Spot Monitoring and Vehicle To Vehicle Communication.

High level architecture diagram: Driverless Car Control Module

I envision the following areas of an autonomous vehicle algorithm need enhancement, customization and adaptation to enable driverless cars in India:

Human-Car Interface

• visual and auditory support for local language instructions
• localized graphical representation of informational & warning signs
• subjects primarily rely on landmarks and not absolute North, South, East, West directional navigation or addresses and street names
• traffic symbols and language per local norms

Route Planning Module

• improved directory to find a location of interest easily based on landmarks and irregular addresses
• real-time traffic updates for better routing
• road construction & maintenance activity updates
• road accessibility information:
  • width / height restrictions
  • speed limits (min, max)
  • type of road (kacha / pakka, rural / urban / motorway)
  • common use or restricted use (bullocks, cyclists, two wheelers, restricted access, trucks, animal crossing, one way)
  • areas to avoid (safety, civil or political unrest, dharnas, bundhs, protests, weather)
  • traffic obstructions (e.g. place of worship located on road, pedestrians on road due to eateries or retail shopping, bazaars)
    

    Traffic in Bazaar. Photo credit: http://commons.wikimedia.org/wiki/File:Traffic_view_from_Charminar.JPG

  • weather patterns that cause driving difficulties (e.g. areas subject to frequent flooding)
  • peak and non-peak traffic timings
  • broken roads, open drainages

In countries like India, all of the above are significant problems till date. While this information is improving with better mapping systems, they need to go a long way before they can deliver top class navigation and routing capabilities.

Environment Perception & Modeling Module

• prediction algorithm has to be robust to detect, track and predict behavior of moving obstacles
• focus on 360 degree surround view with cameras and sensors mounted on front, back, sides and top to provide complete feedback
• ability to process information beyond vision and sensory.
  • impact of horn and sound must be understood to interpret the driver mood and intent: frustration, anger, joy, warning.
  • air pollution levels to understand vehicledensity at the current moment
• use of visual and non-visual cues (e.g. hand signals for turns or stops, hand signals to overtake as is commonly done by bus and truck drivers, flashing lights to indicate a variety of meanings)
• dynamically adjustable tolerance for obstacle proximity detection: considering vehicles drive very close to each other in urban environments, the tolerance limits must be within a few centimeters before precautionary action is initiated by the control module
• support for a variety of vehicles of different shapes, sizes and colours (e.g. auto rickshaws, scooters, mopeds, hand carts, ox carts, cycle rickshaws, tractors, jugaad vehicles)
• support for a variety of life forms and obstacles of different sizes andshapes (e.g. chicken, dogs, pigs, cows, horses, camels, elephants, people sleeping on the streets, piles of trash)
• ability to handle sudden obstacles (e.g. passengers jumping on or off a running bus abruptly)
• variance in safe driving distance depending on the type of vehicle
  • oversized or extra long loads on trucks and buses, without warning signage or lighting
  • buses packed with people including on the roof
  • commuters hanging off transport buses from doors and windows
  • autos, two wheelers and mini-cargo vehicles loaded with people dangerously hanging off vehicle
    

    Passengers dangerously hanging off Bus. Photo credit: www.funnyjunk.com

• ability to handle un-marked barriers
  • potholes
  • speed breakers and road dividers dangerously placed on highways by police
  • crops and grain dried on the roads during harvest season
  • sand, gravel and tar left by road repair crews without signage or safety precautions
  • accident vehicles left dangerously on the road
• support for overcome poor lighting on the vehicle and on the road system
  • vehicles with single headlight, no headlight or taillight, no day time running lights or astonishingly bright head lights
  • excellent heat signature detection and night vision capabilities is required to overcome poor road system lighting
• support to detect traffic violations inspite of signage (e.g. vehicle crossing a red light – as happens between 11PM – 8AM commonly in India)
• Blind spot monitoring – concept of blind spot is not understood in India
• ability to handle non-standardized, multi-lingual and excessively verbose traffic signage

The key take away is that this module must explicitly account the uncertain and incomplete nature of the data and the environment in order to make the right decisions.

Control Module

Based on the highly uncertain and incomplete nature of the data and the environment, it appears that higher frequency coordination and feedback loop is required between the control module and the other modules.

A premise…

Considering the critical nature of the driverless systems, it is absolutely important that they are robust to handle any situation automatically or intelligently pass control to an already distracted driver. A dog running astray, an un-supervised wheelchair or stroller rolling off or a car violating lanes could happen as much in the west as in India. Hence, although these systems are currently being developed in the west where road and driving conditions are organized, I premise that it is important that these systems, out-of-the-box, should be able to handle an environment like India. Else, they just aren’t robust enough and not ready for production deployment in any part of the world. It’s one thing for a reckless or distracted driver to cause an accident. Imagine failure of logic programmed by a corporation – opens up a pandora’s box much larger than the Concorde story.

Driverless cars – boon or bane for India and its future?

I speculate that for driverless cars to gain ground in India, it will take at least over a decade. Substantial change in regulation, increase in awareness amongst all stake holders, indigenization of technology, cost reduction, development of compliance guidelines and testing procedures, standardization and improvement of roads infrastructure, improvement in support technology like reliable mapping & GPS and commercial viability are key to success of the concept on a wider scale.

Relative to developed nations, in developing nations, the per car occupancy is substantially higher, vehicular density on the road is higher and roads much narrower. Consequently, the extent of potential damage to life and property due to failure of a driverless system is substantially higher.

Whenever the driverless car becomes a reality in countries like India, I believe it will be boon. All the challenges – low tech cars, non-tech savvy drivers, high numbers of differently-abled people, poor driver education systems, poor driving discipline, lack of strict driver certification and re-certification process, extremely poor enforcement of traffic rules, increasing number of inexperienced, first-time drivers due to high volumes of new car sales – these are also the reasons that a driverless car could spell opportunity for such markets and bring some order to the chaos.

I can’t wait to get my hands dirty on a driverless car. 🙂

NOTE: If you would like to contribute to this research or have relevant skills and experience to facilitate progress in the open domain on the driverless initiative, please contact me. This article may not be reproduced or republished without express written permission from the author.