Ghost Locomotion, which develops highway self-driving and crash prevention technology, raises $100M Series D led by Sutter Hill (Aria Alamalhodaei/TechCrunch)

100m series sutter hillalamalhodaeitechcrunch

Autonomous driving system developer Ghost Locomotion has raised a $100 million Series D funding round, led by Sutter Hill Ventures. Returning investor Founders Fund also participated in the round, along with Coatue. The money will be used toward R&D as the company continues to develop its highway self-driving and crash prevention technology.

Ghost has been working on a universal collision avoidance technology. The system is premised on the idea that an autonomous driving system doesn’t need to recognize and categorize objects prior to avoiding them – a major paradigm shift. Most systems begin by identifying an object and then use image localization to determine its size, distance and other relevant features.

“We skip that step,” Ghost CEO John Hayes told TechCrunch. “We’re going to recognize anything, any mass that appears in the scene, and then we can get a distance and relative velocity to that. We can start making decisions directly off that data before we’ve classified anything.”

Ghost instead tracks the movement of clusters of pixels in a scene. Hayes pointed to instances where an object is misclassified, or objects that the system hasn’t trained on, as possible failure modes and a reason why classification does not need to be a prerequisite for collision avoidance. Much of this comes down to the certainty of the system’s decisions. According to Ghost, an autonomous system that starts from image recognition is one loaded with lots of opportunity for uncertainty – and it argues, less safe actions on the road.

One obvious counter-argument is that objects should be classified because they behave differently – a vehicle acts differently than a pedestrian, so classification is what allows a system to predict their behavior. But Hayes said that one shouldn’t start with classification, but collision avoidance. “And then if you want to make predictions, you can still do classification,” he said.

One benefit of its system, according to Ghost, is that it requires less computational power – an important consideration for vehicle owners, as higher processing demands can translate into less fuel efficiency. It’s also important for battery electric vehicles that have autonomous driving systems, as each watt of computer power demanded by the AV system can cause a reduction in driving range, Hayes said.

Leave a Reply

Your email address will not be published. Required fields are marked *

Related Posts