|
|
|
|
|
Autonomous Vehicles Still Decades Away: 2019
Analysis Lab |
|
Authored By:Raaga Kannan Thayer School of Engineering Dartmouth College NH, USA Ronald C. Lasky, Ph.D., P.E. Indium Corporation, Dartmouth College NY, USA SummaryThe automotive industry, once predicting the arrival of fully autonomous vehicles by 2020, has backed down from such optimism as industry experts recognize the difficulties of bringing level 5 automation into the hands of consumers. This paper will first introduce the stages of automation defined by the Society of Automotive Engineers (SAE International). Secondly, it will examine the challenges required to progress from existing advanced driver-assistance systems (ADAS) to level 4 and 5 autonomous vehicles. This section will focus on the need for higher precision sensors and software standards as well as the development of cognitive functions such as perception in existing software to navigate daily traffic patterns encountered by human drivers. The inability of current AI technologies to accomplish such a feat will then be discussed. Next, assuming automakers successfully develop the necessary technologies for autonomous vehicles, difficulties of testing the safety of such vehicles will be addressed. This paper will conclude with a discussion of the dangers of releasing level 3 autopilot systems to consumers. ConclusionsIndividuals must look beyond bold articles titled “Autonomous vehicles to carry you to your destination in 2020” to see the fine print stating “exclusions apply to paths containing snow, left-turns, emergency vehicles, construction sites, roundabouts, and pedestrians.” While several auto manufacturers boasted for years the imminent arrival of level 5 autonomous vehicles, industry leaders are recognizing the challenges of full autonomy and scaling back forecasts. Although challenges such as the need for increased precision of electronics and software pose some setbacks, the main issue with level 5 autonomy is the shortcomings of current AI technologies. Lacking the years of context that humans have gathered over their lives, as well as the ability to learn and apply context to new situations as humans so easily do, engineers struggle to design autonomous vehicles that demonstrate the adaptability that make humans good drivers. Given the high variability of driving, finding sufficient methods around teaching a computer cognitive functions have proven difficult. Current inductive learning methods pose several limitations given the prevalence of blind spots and the difficulty of identifying these blind spots. Thus, designing level 5 autonomous vehicles that respond appropriately to all edge cases imaginable is an impossible task. Not only is designing such a vehicle extremely difficult, testing the safety of the vehicle to ensure it meets all standards and performs as well as human drivers is another great challenge. Methods for testing, including augmented reality environments and computers running virtual mileage, rely on a library of artificial edge cases that once again runs into the issue of identifying and correcting all blind spots. With great difficulties in designing and testing level five systems, fully autonomous vehicles are several decades away. Although the technology for level 3 systems are more present, the dangers of these systems may outweigh the benefits and have caused many companies to move away from implementation. Thus, aside from Tesla which hopes to achieve full autonomy from improvements to its level 3 system, we are likely to see level 4 systems from competitors arrive next to the market. Although less impressive than flashy headlines describing robo-taxis, level 4 autonomy still promises benefits to consumers. Meanwhile, the advancements ahead in AI and electronics to make fully autonomous vehicles a possibility are exciting and stand to benefit society in several other facets. Initially Published in the SMTA Proceedings |
|
Comments
|
|
|
|
|
|
|
|
|