The need to socially distance during the COVID-19 pandemic is causing many autonomous vehicle manufacturers to slam the brakes on automated taxi services. However, the pandemic has created new opportunities for companies to shift their focus to transporting food and goods via delivery services provided in methods ranging from autonomous trucks, vans, and drones, to sidewalk delivery robots.
“The reality right now is that goods delivery is a bigger market than moving people,” John Krafcik, CEO of Google’s Waymo, told Reuters.
Even before the pandemic, many autonomous vehicle manufacturers already had their sights set on autonomous delivery services. Now, investment in these services is accelerating. In the last seven months, over twenty autonomous vehicle makers received roughly $6 billion USD to develop delivery systems, according to a recent Reuters report. Since the start of 2020, the majority of that investment—over $4 billion USD—has been made to U.S.-based group Waymo and Chinese transportation company Didi Chuxing. Both companies are making autonomous vehicles for both passengers and deliveries.
How Simulation Technology Can Speed Autonomous Vehicle Training
Before they can safely be deployed on roads, autonomous vehicles need eleven billion miles worth of on-the-road training. To hasten this process, some manufacturers plan to use AI-generated simulations that can train vehicles much faster than on-the-road testing.
In May, researchers from Waymo published a paper on how artificial intelligence can be used to create highly accurate training simulations that mimic real roadway scenes. Dubbed “SurfelGAN”, the technique employs texture-mapped surface elements to recreate realistic roadway scenes. This is done by using data previously captured on cameras and lidar sensors outfitted on Waymo’s self-driving vehicle fleet. As a whole, the fleet has collectively criss-crossed over 20 million miles of public roads.
While there are different ways to create simulations, the researchers claim SurfelGAN presents a simpler way to simulate sensor data.
“We’ve developed a new approach that allows us to generate realistic camera images for simulation directly using sensor data collected by a self-driving vehicle,” a Waymo spokesperson told VentureBeat. “In simulation, when a trajectory of a self-driving car and other agents (e.g. other cars, cyclists, and pedestrians) changes, the system generates realistic visual sensor data that helps us model the scene in the updated environment … Parts of the system are in production.”
The system also includes a generative adversarial network (GAN) module that allows it to create authentic-looking images for training simulations.
While the researchers acknowledged some flaws in the system (which can occasionally result in abnormal-looking renditions of cars, for example), they claim SurfelGAN is a solid contender for training simulations.
“Simulation is a vital tool in the advancement of self-driving technology that allows us to pick and replay the most interesting and complex scenarios from our over 20 million autonomous miles on public roads,” a Waymo spokesperson told Venture Beat. “In such scenarios, the ability to accurately simulate the vehicle sensors [using methods like SurfelGAN] is very important.”
In March, researchers from MIT also announced they created a system that uses simulations to train autonomous vehicles called “Virtual Image Synthesis and Transformation for Autonomy” or VISTA. Like SurfelGAN, VISTA employs real-life data collected from road tests that allow it to recreate photorealistic simulations. It employs a reward system that encourages vehicles to continuously improve. When a vehicle’s controller successfully travels a specific distance within VISTA without getting into an accident, it’s given a reward. This incentivizes the controller to learn from its mistakes.
“It’s tough to collect data in these edge cases that humans don’t experience on the road,” Alexander Amini, a PhD student in the Computer Science and Artificial Intelligence Laboratory (CSAIL), told MIT News. “In our simulation, however, control systems can experience those situations, learn for themselves to recover from them, and remain robust when deployed onto vehicles in the real world.”
Prepare for an Autonomous Future
Prepare your organization for the latest developments in AV technology with training in foundational and practical applications. Developed by leading experts in the field, the IEEE Guide to Autonomous Vehicle Technology is an online seven-course training program.
Connect with an IEEE Content Specialist today to learn more about purchasing the program for your organization.
Interested in purchasing the program for yourself? Access it through the IEEE Learning Network (ILN)!
Wiggers, Kyle. (20 May 2020). Waymo us using AI to simulate autonomous vehicle camera data. Venture Beat.
Lanhee Lee, Jane and Lienert, Paul. (18 May 2020). Automated delivery cashes in on pandemic-driven demand. Reuters.
Mathewson, Rob. (23 March 2020). System trains driverless cars in simulation before they hit the road. MIT News.