Virtual Driverless Tech Testing Not a Stand-Alone Solution

Digital simulated testing for autonomous vehicles (AV) has been seen as a cheap, risk free, alternative to testing in the real world.

Yet, while exponents voice its many advantages, still more voices are raised against trusting virtual world results and conclusions. “Real-world testing requires a great deal of time and investment, and legislation requires permits for testing of autonomous vehicles on the road,” Sam Barker, senior analyst with Juniper, told TU-Automotive. “Using simulation software enables the developer to mitigate this, as well as control the testing environment and minimize the risk from real world testing.”

Barker also noted that these systems usually employ artificial intelligence (AI), which, in turn, could feed back results of testing to real world environments in the future. He explained all AV systems will also need to be put through rigorous real world testing before any commercial launch occurs and improvements in a virtual test platform’s capabilities and accuracy of results is likely to come from the data fed back from real world testing.

“We can look at software platforms as an initial proving ground before systems are tested in real world environments,” he said. “As the capabilities of these platforms improves, this will lessen the dependence on real world testing.” Barker also claimed testing on platforms will also be more cost efficient and able to scale, meaning that more testing can be done over these simulation-based platforms before moving on to real world testing on roads.

Toyota AI Ventures, the venture capital subsidiary of the Toyota Research Institute (TRI), has invested in Parallel Domain, a Silicon Valley-based software developer specializing in 3D environment generation for autonomous vehicle simulation.  The company’s CEO Kevin McNamara said one of the biggest limitations of these platforms is content. “Building the environments and scenarios required for large scale simulation is a huge bottleneck for developers right now,” he explained, noting most AV developers don’t simulate nearly as many miles as they’d like to because they just don’t have those miles and scenarios built into their simulation. “Further, they don’t have automated ways to translate what their cars see in the real world into simulation,” McNamara said. “They’re racing to increase this mileage and that’s one of the big reasons we’re focused on solving this problem for them.”

However, with simulation AV developers can drive more virtual miles and capture more images at a scale many times larger than they could with real world operations. “In an industry where data is king, this is a huge differentiator,” McNamara said. “The maturity of a company’s simulation team will be one of the primary variables in deciding who will win this race.”

Generally speaking, AV developers need simulation software to become much more flexible and robust to handle a drastically wider variety of environments and situations while becoming easier to use, deploy and scale. “Lowering this barrier to entry while simultaneously maturing the feature sets available will be the key to success for simulation providers,” McNamara said.

He explained that while automotive simulation has been around for decades, the idea of simulating a Level 5 autonomous vehicle is significantly more complex, especially given the addition of sensors and perception. This makes for a new software ecosystem with a lot of different companies, formats and standards popping up in an ad hoc manner. “I think the next few years will see significant maturation in the different simulation offerings, potentially some consolidation and, hopefully, a lot more interoperability between packages,” he said.

Arizona, one of the nation’s most ambitious states for autonomous vehicle development and testing has enlisted tech giant Intel to help with a simulation lab at the Institute for Automated Mobility (IAM), to be built alongside a 2.1-mile test track. “We will see new entrants into the market if they can feel they can offer more comprehensive testing abilities,” Baker said.

He noted, however, that the addressable base of potential users is low and limited to automakers and others in development of autonomous systems, which would probably discourage any potential new entrants. At the same time, Baker said he expects carmakers to continue to use third parties for these systems, owing to the high investment costs and the ability to benefit from their expertise in AI.

McNamara noted partnerships, like the one between Toyota and Parallel, are critical to help large organizations stay on top of nascent technology. “We will see some very big players getting into the simulation business during the next few years,” he said. “We’re still many, many years away from ubiquitous Level 5 autonomy and the demand for simulation is going to increase significantly as companies embark on the ‘long tail’ of moving their technology from proof of concept to production ready.”