Recogni is a company that designs and develops an AI-based visual perception system for autonomous vehicles.
We thank Ashwini Choudhary for sharing several insights from the world of AI in the automotive industry, including:
- The need for extremely high computational power for the energy budget of a self-driven vehicle and how Recogni is tackling this challenge.
- The progress of the automotive industry towards making safer self-driving cars.
- The trends emerging in the autonomous vehicles industry, and how AI advancements in this field are bringing new business opportunities.
What is your background and how was Recogni started?
I am a serial entrepreneur, having founded multiple successful companies in the semiconductor and systems industries. Back in 2013, I suffered a life-changing motorcycle accident which posed the question: Is there a way for vehicles to automatically avert collisions and alleviate dangerous situations such as the one I went through?
After months of rehabilitation and surgeries, I chatted about this with a former colleague and a classmate of mine. Delving into this idea, we discovered that there is nothing today to enable cars to be autonomous and eliminate or mitigate situations such as the one I endured. This problem we discovered led to the founding of Recogni in 2017.
What is Recogni’s mission?
Our mission is to design a vision-perception platform purpose-built for autonomous vehicles to flawlessly perceive their surroundings, enabling them to drive safely and efficiently.
Tell us about Recogni’s solution and what specific problems it solves.
We are the only solution on the market that has the efficiency to solve the visual perception problem faced by the autonomous vehicle industry.
To correctly interpret visual cues in a few milliseconds while driving, your brain uses 86 billion neurons to compute 10000 Tera-Operations-Per-Second (TOPS), all while consuming a few watts of power. Clearly, for a solution to be optimal for autonomous driving, it would need extremely high processing power while also consuming a miniscule power envelope (especially important, considering the trend of vehicle electrification). This unsolved optimization problem highlights the largest barrier to making fully autonomous vehicles a reality: Visual Perception Problem
Recogni’s platform is uniquely purpose-built to solve the visual perception problem. By are leveraging key innovations in algorithms, architecture, and ASIC implementation, and uniquely applying them, we facilitate our product to have 1000 TOPS of processing power, while consuming a mere 10 watts. These unmatched capabilities enable an AV to see a traffic light 200 meters away and correctly interpret visual cues very quickly.
Essentially, Recogni is the only solution on the market today that can make fully autonomous vehicles a reality, and we do that by solving the visual perception problem.
How do you position Recogni’s offer in the automotive space? How does Recogni do it differently?
Recogni is the only product on the market capable of achieving the processing requirement necessary to enable full autonomy. Thus, we are the only ones that can enable auto manufacturers to both develop partially autonomous vehicles (AVs) today and scale to full autonomy in the future.
Right now, the auto industry is facing two evolutions: a transition to vehicle autonomy and electrification. Given the 5 levels of vehicle autonomy, each level has an exponentially higher processing requirement compared to the previous level. Currently, Tesla has a major advantage on traditional car manufacturers in both of these aspects – they are internally developing an AI Vision solution specifically to enable their EVs with autonomy, and this platform has performance capabilities far beyond those of its competitors.
There are a few ways to chase Tesla, including using a solution developed by Mobileye, NVIDIA, or by one of the many accelerators coming into the market. However, none of these products, including even Tesla’s, has the architecture to enable vehicles to scale to full autonomy, and it all comes down to simple math: the biggest roadblock to achieving full autonomy is the fact that over 75 TOPS per watt of efficiency is needed to solve the visual perception problem stated earlier without compromising the battery.
Although Tesla hasn’t completely solved the visual perception problem yet, their significant advantage in developing an AV solution poses a problem for traditional car manufacturers. Because products today cannot meet the processing requirements for full autonomy, companies will have to spend lots of time and money to research and develop autonomous vehicle technology to close the gap between themselves and Tesla. Rather than going through this, these companies want a new, purpose-built platform to save money – something that can solve the full autonomy problem of the future, and inherently also the partial autonomy problem of today.
Our product is purpose-built to solve the visual perception problem, while taking into account electrification – our chip is ultra-efficient with 100 TOPS per watt, and has enormous compute with 1000 total TOPS of processing power on the chip, showing that we are more than capable of meeting the processing requirement to solve the visual perception problem. In addition, in terms of “dollars per TOPS,” our product is orders of magnitude cheaper than anything on the market, demonstrating unmatched scalability and affordability.
Long story short, our product is the only solution on the market today that can enable full vehicle autonomy. We can enable car manufacturers to gain a competitive advantage over Tesla by equipping them with the technology needed to enable partial autonomy today as well as full autonomy in the future. Incumbent solutions are unable to do this; thus, they are set to pose a significant negative burden on auto OEMs in the future as the industry evolves to vehicle autonomy and electrification.
What has been your biggest challenge so far?
Our biggest challenge is that we are the new guys on the block.
There are already industry incumbents backed by Fortune 100 corporations who are trying to solve the visual perception problem to make autonomous vehicles a reality. However, we believe we can overcome this because we are confident that our product’s technological and performance advantages will push auto manufactures to integrate our solution.
Tell us about the Recogni team.
We have a cross-functional team of experts in the auto, AI, sensor, and semiconductor industries. We also have a management team with a proven track record of bringing game-changing products to the market and overseeing the successful exit of countless startups.
How far has the autonomous vehicles industry come in making self-driving cars safer than human-driven ones?
Today, the autonomous vehicles industry has a long way to go in making self-driving cars safer than human-driven ones, as there is no technology that can enable full autonomy, thus eliminating the non-zero human reaction time.
The human reaction time is detrimental to drivers in accident situations. Because of this, by the time the driver makes a maneuver to avoid the accident, it is often too late. For self-driving cars to be safer than human driven vehicles, they need a solution to eliminate the non-zero human reaction time. By doing that, the car will have ample time to create a safe path plan to avert deadly situations such as fatal head-on crashes and collisions with vulnerable road users such as cyclists and pedestrians.
Solutions today cannot eliminate the human reaction time by enabling full vehicle autonomy– they can only enable partial autonomy, which requires a human driver to make the necessary maneuver. For the safety benefits of autonomous vehicles to be realized, a new purpose-built solution must be developed capable of detecting, interpreting, and relaying visual cues in just a few milliseconds, enabling cars to drive on their own in a safer manner than human-driven vehicles today.
What trends do you see emerging in the autonomous vehicles industry?
The autonomous vehicles industry and its respective intersections between the logistics and ridesharing industries are interesting trends to me.
Last-mile and local delivery companies face high operating costs due to human drivers. However, autonomous vehicles serve as a solution to replace human-driven vehicles and increase profit margins for these companies, allowing them to be sustainable in the long run.
In addition to last-mile, autonomous vehicles bring benefits to long-haul delivery. Given both the fact that there exists a trucker shortage, logistics companies need to find a solution to meet the ever-growing demand of delivery. Not only can autonomous vehicles mitigate this issue, but they bring fuel-saving and safety efficiency benefits, both of which increase profit margins.
Delving into ridesharing, Uber and Lyft both invest millions in developing self-driving technology for their vehicles, and the reason is the same as that for last-mile and local delivery – human labor places a huge burden on these companies’ operating expenses. Therefore, they are looking for a robo-taxi solution to replace their current human-driver model and increase their profit margins.
The mission of AI Time Journal is to divulge information
and knowledge about Artificial Intelligence, the changes that
are coming and new opportunities to use AI technology to
Are you interested in sharing information and knowledge about AI and connecting with some of the brightest minds in the field?
Learn about joining as an Associate Editor.