Connected and autonomous vehicles (CAVs) have a future.
That is without doubt but there is still a need to ensure that they will be safe on our highways and to ease the public’s safety concerns to increase their adoption over the next few years. CAVs need to be able to react to unforeseen events – just as we do as human drivers.
None of the obstacles that drivers face can be anticipated and so the artificial intelligence and machine learning of these vehicles needs to be able to react accordingly to prevent accidents from occurring, no matter the weather, no matter the circumstances, and not matter what kind of environment – both rural urban – they are to operate in.
Edge case scenarios
Danny Shapiro, senior director automotive at Nvidia, nevertheless believes that “autonomous driving in normal situations is a solved problem.” Edge cases are what makes autonomous driving hard, he says while referring to the “rare and difficult situations that often include a combination of challenges.” This might be, for example, someone accidentally losing control of a shopping trolley that rolls onto the highway. It’s not a common event, but it could happen.
He adds: “Then there’s the added complexity of making sure the car can address compounding conditions, such as having this occur at night, in the rain, and with an additional driver in front of you who blocks the view of the cart until the last minute.”
Alice Salter writes for 2025AD in her article, Are artificial cities the answer to AV road safety?: “While you could drive for days without encountering some of even the most common obstacles you’ll find on the road, in controlled environments like artificial test cities, technicians can guarantee them. This proves vital in forming public opinion of, and trust in, the tech. While you could drive for days without encountering some of even the most common obstacles you’ll find on the road, in controlled environments like artificial test cities, technicians can guarantee them.”
Shapiro responds: “In order to ensure a car is truly safe, it requires testing these situations, and a wide range of permutations, repeatedly as part of development and validation. In the real world, this situation is rarely encountered in on-road testing and certainly never happens exactly the same way twice. Therefore, we need an alternative to ensure the car knows how to properly respond.”
“Simulation in an artificial city (that mirrors reality) provides a way to test the same situations over and over. It also allows us to add variations in a controlled way. This means we can help speed progress since we spend time testing the hard things and, because it’s repeatable, we can measure progress as we work.”
Maximising connectivity
Stéphane Barbier is chief business development officer at Transpolis SAS, which describes itself as being “the unique smart city lab in Europe dedicated to innovative transportation systems and road equipment”. He explains that one scenario for the use of artificial cities for CAV development could see their design and construction being optimised to have “maximum of connected devices and intelligent infrastructures, in order to have the best connectivity for CAV.”
Cities could be developed to host only CAVs. This would ease the traffic. He argues that it’s too complex to have CAVs in crowded areas, sharing with a mixture of traffic types – include space where non-automated vehicles are driven, and where there are cyclist and pedestrians going about their day. However, the reality is that the city traffic of the future is likely to involve a range of different vehicle types, and so artificial cities should ideally be used to reflect this situation.
Barbier recognises the limitations of developing artificial cities for CAVs only. “Developing artificial cities for CAV only is not the best way to design and build a city, which must be created for its inhabitants and not for the transportation system only.”, he comments before suggesting that “living in a world of CAV only would be something strange and really unusual.”
Different levels of detail
Shapiro therefore thinks that artificial cities can be design and built with “different levels of detail and accuracy”. So, what is the biggest challenge? Building detailed artificial cities is labour intensive. “Getting the right variety and level of ‘noise’ and realism is challenging”, he says. While simulation is a model of the real world, the accuracy of the data emanating from it has to be tested to ensure any findings are evidence-based and accurate, reflecting real-life scenarios and operations.
He explains: “No model is perfect but they can be accurate enough to provide real answers. In order to assess how well the models work and to provide confidence in the results, the models need to be compared against real-world data. The goal of simulation is not to eliminate real-world testing, but to make real-world testing more efficient.
“AI-powered autonomous vehicles must be able to respond properly to incredibly diverse situations on the road, such as emergency vehicles, pedestrians, poor weather conditions, and a virtually infinite number of other obstacles, including scenarios that are too dangerous to test in the real world.”
Highway testing
Simulations are a crucial tool. At present in most jurisdictions, road testing on actual highways is not feasible in all of the potential situations that might arise while driving on the highway. Shapiro adds that road testing is not “sufficiently controllable, repeatable, exhaustive, or fast enough, and that’s why testing and training in simulation is so important because it allows us to test all of these possibilities in the virtual world.”
Barbier underlines that simulation is about safety and that no automaker would ever want to sell vehicles that have not been 100% tested. Simulation and track testing is essential before the vehicles can be driven on city streets. “Real-life scenarios have to be taken into account in simulation, in controlled environment and in the public space with the necessary safety conditions,” he explains.
He also believes it’s not a weakness to test and analyse potential scenarios using artificial cities and simulations – pointing out that it involves a complete value chain from simulation to testing in a controlled environment and eventually on the open road. Testing in rea-life environments, he warns, would be too dangerous.
Physical testing: invaluable
Shapiro concurs that there are many advantages with testing in a controlled environment. It makes it possible to test the more difficult rare and dangerous situations that an autonomous vehicle might find itself in on the road – enabling testing and analysis without putting anyone, or perhaps anything in the case of autonomous vehicles, in harm’s way. He still finds that physical testing is invaluable too. It creates an opportunity to collate critical training data, helping developers to “continue to refine their software algorithms to ensure safe autonomous driving.”
He concludes by saying that artificial cities, “can best be used as a controllable, repeatable, well-understood test environment to develop and test autonomous vehicles.” They will permit autonomous vehicles to achieve the level of experience they need to negotiate rate and dangerous situations. Artificial cities therefore seem to be a great way forward, helping to improve CAV development by making sure that connected and autonomous vehicles are safe to operate.
by Graham Jarvis
Source: https://www.tu-auto.com
CUT COTS OF THE FLEET WITH OUR AUDIT PROGRAM
The audit is a key tool to know the overall status and provide the analysis, the assessment, the advice, the suggestions and the actions to take in order to cut costs and increase the efficiency and efficacy of the fleet. We propose the following fleet management audit.
