Auto Industry can take a leap forward in autonomy and safety with SiMa.ai

Harald Kroeger, a veteran auto tech executive, believes that one day in the very near future, traffic incidents will be a thing of the past. Sure, we all dream of being able to watch the latest episode of the “Morning Show” as we fly down the interstate. And autonomous taxis grab all the headlines. But the real goal should be to completely eliminate human harm from traffic incidents. He sees that reality coming sooner rather than later.

The automotive industry has a long history of emphasizing the development of next-generation safety features to protect consumers. It’s been integral to the overall design of cars, from crash protection such as crumple zones to structural innovations like high-strength steel. Early innovations included the windshield wiper in 1903, safety glass in the 1920s and hydraulic brakes in the 1930s. 1959 brought the three-point seatbelt, and by 1968 seat belts were mandatory in new U.S. cars. In 1971 anti-lock brakes arrived and in 1974 airbag regulations emerged.

Still, there are more than one million auto-related deaths worldwide every year—a statistic so dire that the United Nations has made it a priority to cut global road traffic deaths and injuries in half by 2030. While the introduction of many features has improved safety for passengers over the decades, it’s contemporary AI-based technologies that hold the promise of finally reducing harm from traffic incidents by several orders of magnitude. It is now possible to envision a world where there are essentially zero fatalities from auto accidents, said Kroeger, who is president of Automotive forSiMa.ai, a software-centric company that has built a multi-modal edge AI platform that optimizes for power efficiency, responsiveness, and reliability –all attractive attributes for automotive use.

 “My vision is the first step will be to make sure cars are better equipped to contribute to the safety of the people that ride in them. If we can reach this next level of autonomous vehicles, cars can have enough ‘brains’, sensors, and ‘understanding’ so that they can prevent something blatantly wrong, such as swaying into the wrong lane, going off the road, or hitting a tree or another motorist,” Kroeger said.

The auto industry can make this happen without having to achieve full automation. In fact, Kroeger would argue that full,L5-level autonomy shouldn’t even be the near-term goal. Instead, improving a concert of current L1 - L2+ features—such as lane assist, blind spot detection, automated emergency braking, forward collision warning, pedestrian detection and autosteer—ought to be the auto industry’s focus.  

 These safety features need to be trusted, reliable, and more proactive than they are today. If a system nags too much, it’s likely to be disabled. And if it’s too hard to understand, it’ll be ignored. Safety features also need to be better at working in a variety of environments including at nighttime, during inclement weather, through more challenging traffic scenarios, or during temporary changes such as construction. And while alert systems are effective, systems that take action such as autosteer and automatic braking are even better at avoiding or lessening the impact of a collision.

Despite the effectiveness of these systems, getting to this level of safety sophistication up and down every major make and model of car is going to require significantly more compute at the edge.

New Infrastructure Required

Many of the major leaps forward in safety are being driven by new technology's use of powerful machine learning algorithms. These algorithms can ingest a variety of inputs from the car’s sensors and then make split-second automated decisions—augmenting human capability.

But these systems have a number of requirements in order to be viable at scale:

• The system must make decisions with very low latency, in the blink of an eye, to avoid accidents. This critical safety factor is different from the requirements of, say, answering a prompt on a computer or mobile phone. As such, this can require heavy computation to make the correct decision extremely fast.
• This must be done with minimal energy use to not sap the car’s battery life and to not require additional cooling.
• Finally, this must all be done in a manner that enables affordability so that these features can be broadly accessible beyond higher-end makes and models.

Today's self-driving cars tend to be at odds with electric vehicles as well. Autonomous vehicles rely on the large sensors, from cameras, to radars and LiDARs, bolted across the vehicle to collect, process and analyze infinite amounts of data, which in tandem require massive amounts of energy to simultaneously power and cool down.

This is a solution that SiMa.ai has been working on for the past six years. The company’s software-centric Machine Learning System-on-Chip (MLSoC) provides performant machine learning at the edge. And with its unmatched power efficiency and extremely strong software backbone, it can provide the missing piece in the puzzle for auto safety.

Autonomous vehicles go hand-in-hand with the edge. SiMa.ai purpose-built modular software and hardware for the embedded edge to allow AI to understand and respond to the physical world. Further, when a vehicle’s AI is processed on its own edge environment, energy consumption is better optimized in electric cars. Installing the car’s AI brain into the vehicle, rather than making it connect to a distant data warehouse, is the beginning of a future where roads are safer than ever before.

‍Auto Manufacturers Leading the Way

“I’ve been in the automotive business for the better part of my career and have worked with some of the most innovative companies in the space,” said Kroeger. “Every major manufacturer is looking for solutions to build safer cars. It’s not just about regulatory compliance or sales slogans. They care deeply about the safety of their drivers. SiMa.ai will be the platform that gets them there .”

‍