How to Teach Your Car to be Autonomous
14 June, 2017
Cognata Raised $5 Million from investors, including Airbus Ventures, to launch innovative Deep Learning Simulation for the speeding of Autonomous Vehicles development
Bringing an autonomous vehicle to the road is not an easy task. Beside the huge technical challenges, the final product must be tested, evaluated, re-tested, re-designed, re-evaluated and receive all the needed approvals. According to Rand Corp such vehicles will need to drive 10 billion miles in order to reach a human-level error rate —and that can take years to achieve. Rand: “Autonomous vehicles would have to be driven hundreds of millions of miles and sometimes hundreds of billions of miles to demonstrate their reliability in terms of fatalities and injuries.”
But the industry do not ready for this: Even the largest developers of autonomous vehicles have only accumulated a few million miles in autonomous mode after years of testing. Cognata, Ltd., a startup company from Rehovot, Israel, has came up with a smart solution to this problem: artificial intelligence, deep learning, and computer vision in a simulation platform that enables autonomous vehicle developers to shave years off the long, costly process of road-testing autonomous vehicles.
Today Cognata announced the launch of its simulation engine, supported by $5 million in funding from Emerge, Maniv Mobility, and Airbus Ventures. The company will use the funding to accelerate product development and commercialization of its new solution.
Realistic Driving Environment
Cognata uses patented computer vision and deep learning algorithms to automatically generate a whole city-simulator including buildings, roads , lane marks, traffic signs and even trees and bushes. The core of the the product is a deep learning simulation engine leverages reality-grade city mesh combined with DNN (deep neural network) and AI capabilities.
It adds dynamic layer of realistic traffic model of other vehicles and pedestrians. Historic local weather conditions and lighting are added to stress test the system. The simulation engine also simulate the sensor interaction with the external materials including sensors ,saturation and lost packets to receive the most comprehensive autonomous driving simulation feedback loop from each drive.
The simulation engine reproduces sensor input in high fidelity by emulating interactions with real-world materials. Furthermore, Cognata can recreate cities anywhere in the world, allowing a dramatically expanded range of testing scenarios beyond the current limited geographies, to the great benefit of any OEM and Tier-1 autonomous vehicle manufacturer.
Cognata was founded in 2016 by a team of experts in deep learning, autonomous vehicles and computer vision. The CEO, Danny Atsmon, is a deep-learning expert who formerly served as Harman’s director of ADAS. He explained: “Our simulation platform rapidly pumps out large volumes of rich training data to fuel these algorithms. We introduce a scalable, safe, and cost-effective solution that not only accelerates the development and validation of autonomous vehicles, but also ensures that vehicles can be much safer when they’re deployed in the real world.”
François Auque of Airbus Ventures adds, “After extensive technical evaluation of the team and the technology, we look forward to using Cognata’s technology, which can accelerate the time to market for land-based autonomous systems as well as the kind of aerial systems we are focusing on.” The current investors in the company includes Emerge venture capital firm, Maniv Mobility, Israel’s first venture capital fund dedicated exclusively to the new mobility future, and Airbus Ventures, headquartered in Silicon Valley and Paris.