Recogni raises $48.9 million to develop AI perception chips for autonomous driving

Recogni is a startup that designs AI visual recognition models for self-driving cars. According to foreign media reports, the company recently announced that it has received $48.9 million in financing and said it will use the funds to bring its perception products to market while expanding the size of its engineering and market development teams.

(Image source: Recogni)

The company's CEO RK Anand believes that self-driving cars have a computing problem. Although the models that guide the car's decisions have powerful training servers, inference, the stage where the algorithm makes predictions , must be performed offline to ensure redundancy. However, even well-trained models require on-board computers for real-time processing, which some people believe is unsustainable.

Recogni’s integrated model includes a passively cooled image sensor, an external depth sensor, and a custom inference chip that can perform 1 peta operations per second while consuming only about 8 watts of power. Because the chip offloads central processing tasks to multiple points in the vehicle, it can capture and analyze up to three uncompressed 8-12MP streams at 60 frames per second, achieving 70% computational efficiency in typical vision applications.

Recogni claims that its system outperforms competing products by more than two orders of magnitude in perceptual tasks such as image classification, object detection, action prediction , and deep inference. In the benchmark ResNet 50, the model can classify 92,105 images per second; in RetinaNet-101-800, it can perform 1,750 inferences per second; and in R(2+1)D, it can recognize 833 people at the same time.

A company spokesperson said, “Recogni’s AI visual cognition model is designed to capture and process high-resolution camera and sensor data. Through innovations in AI algorithms, ASIC architecture and design, and system software, the model will provide high-performance processing performance with ultra-low power consumption, enabling simultaneous real-time processing of high-resolution, high-frame rate images from multiple cameras.”

Recogni plans to first target L2 autonomous vehicles, including cars equipped with ADAS such as Cadillac's Super Cruise, Nvidia's Drive AutoPilot, and Volvo's Pilot Assist system. In the future, the company plans to build a platform for L3 and L4 autonomous vehicles, with the ultimate goal of making cars able to drive themselves as well as human drivers.

A strictly vision-based approach to autonomous driving is widely favored, and Intel-owned Mobileye is also a strong supporter of this approach. The company is developing a custom accelerator processor chip that uses patented algorithms, cameras and ultrasound to achieve 360-degree coverage. Self-driving truck startup TuSimple says its camera-based technology can detect up to 1,000 meters. Baidu also recently launched a car framework Apollo Lite that it claims can achieve fully autonomous driving on public roads.

Recogni also competes with Tesla. In April 2019, Tesla detailed a Samsung-made chipset with AI performance that can perform more than 144 tera operations per second. Nvidia, another competitor in the field, said its Drive Xavier chip system at the heart of its Drive AGX Pegasus autonomous vehicle development platform uses just 30 watts of power.

Anand said, “This investment is a strong endorsement of Recogni by venture capitalists and industry leaders. Although Recogni has not yet mass-produced this model, it has completed proof of concept with several automotive customers. We are building the highest performance AI inference system with the lowest energy consumption to solve the challenges of perception processing and energy efficiency.”

WRVI Capital, Mayfield Fund, Continental Automobiles, Robert Bosch Venture Capital, and existing investors GreatPoint Ventures, Toyota AI Ventures, BMW i Ventures, Fluxunit-OSRAM Ventures, DNS Capital, etc. participated in this Series B round. Following the $25 million round in July 2019, the company’s total financing has now exceeded $70 million.