Offering increased productivity, accessibility, efficiency, a more positive impact on the environment, and the potential for additional safety, autonomous vehicles are widely regarded as the future of urban transportation. But how can self-driving cars and other forms of autonomous transport be seamlessly integrated into existing cities' infrastructure and environments?
In contrast to the simplicity of driving on highways, planning and decision making for autonomous driving in cities is much more difficult as it requires progress in both transport technology and infrastructure to effectively deal with the increase in operating velocity of autonomous systems as well as the complexity of urban environments.
At the Machine Intelligence in Autonomous Vehicles Summit
in Amsterdam, we'll explore motion planning for autonomous vehicles in urban areas further with Hans Andersen, a PhD Candidate at the National University of Singapore
. Hans will share expertise on recent advances in this field, model predictive control, and the applications of robotic motion planning to improve autonomous driving.
I spoke to him to learn more ahead of the summit on 28-29 June.How did you begin your work in autonomous vehicles?
I was looking around for PhD opportunities in the field of robotics, and one particular topic that interested me was autonomous driving. Our group started with a single golf buggy in 2010. When I joined in 2014, the group was expanding its capabilities to include more vehicles, and to start testing in more challenging situations. We conducted the first autonomous vehicle public trial in Singapore in October 2014. Today, we have a fleet of five golf buggies, a road car, and two personal mobility vehicles.
What are the key factors that have enabled recent advancements in motion planning for autonomous vehicles in urban environment?
Recent developments in fundamental robotics research such as algorithmic motion planning, especially in the areas of probabilistic motion planning has contributed a lot to the advancements in the capabilities of autonomous vehicles. Another research area that has also contributed to the advancements are intelligent control and optimization methods. The availability of urban driving dataset in the research community and the reduction in cost of computing power especially in parallel computing (GPU), encourage applications of machine learning techniques in both autonomous driving perception and planning. Combined with realistic modeling and simulation of urban driving environment, researchers have access to the tools that enable rapid development of new autonomous driving capabilities, without necessarily having a test vehicle, which can be costly.
What are the key challenges to progress for autonomous driving in cities?
Different cities have different driving characteristics and traffic rules, and therefore what works in one environment, may need a lot of refinements if it’s applied to a different environment . The unpredictable nature of the agents around the autonomous vehicle is still a challenge in planning appropriate actions for the vehicle. capturing intentions of nearby pedestrians and cyclists for example, is still an active research area. Another challenge in autonomous driving is how to make the autonomous vehicle behave more like a human driver, especially in situations where the vehicle has to break some traffic rules, e.g. when overtaking an illegally parked car. I believe that there are still many problems that will appear in hindsight as we clock in more autonomous driving mileage. These new challenges can be identified and overcome if we get access to perform more real world test, and increase not just the mileage, but the complexity and diversity of the environments in which the autonomous vehicles operate.
What developments can we expect to see in autonomous vehicles in the next 5 years?
With autonomy packages that offer semi autonomous driving capabilities coming to cheaper cars, researchers will get access to more data about human factor in autonomous driving, and design future improvements accordingly. Currently, fully autonomous driving tests in cities usually involve at least 2 human operators in a single car: a safety driver and an engineer that monitors the state of the system. I hope that we can see a full deployment of autonomous vehicles in a city without a safety driver as a backup in the near future. There are also a lot of research and testing in smarter, more connected cities, such that the surrounding infrastructure can provide information to the vehicle and vice versa. Hopefully with converging standards and regulations, we can see more connected (autonomous) vehicles operating cooperatively in the near future.
Outside of your field, what area of machine learning advancements excites you most?
I am very intrigued by the application of machine learning in healthcare, especially in diagnostics ass diagnoses can be very prone to human error. I am very excited to see the role of computing in getting faster and more accurate diagnoses for patients in remote areas where the quality of healthcare practitioners and availability of medical equipments can often be inadequate.
Opinions expressed in this interview may not represent the views of RE•WORK. As a result some opinions may even go against the views of RE•WORK but are posted in order to encourage debate and well-rounded knowledge sharing, and to allow alternate views to be presented to our community.
|Hans Andersen will be speaking at the Machine Intelligence in Autonomous Vehicles Summit in Amsterdam on 28-29 June, taking place alongside the Machine Intelligence Summit. View more information here.|
Confirmed speakers include Pablo Puente Guillen, Researcher, Toyota Motors; Jan Erik Solem, Co-founder & CEO, Mapillary; Pejvan Beigui, CTO, EasyMile; Jim Aldon D'Souza, Autonomous Driving Research Engineer, TomTom; and Tarin Ziyaee, Director of AI, Voyage; Julian Togelius, Associate Professor, New York University.
Tickets are limited for this event. Book your place now.