The road to full, functional autonomy

Opportunities and challenges in the autonomous vehicles industry.
Santiago Tenorio is a Venture Partner at Rewired.
This article was written by Sparsh Jain, Intern at Rewired – Summer 2018

In 2018, self-driving vehicle technology is expected to make a significant leap towards achieving full, functional autonomy, the SAE Level 4. The evolution from SAE Level 3 (Conditional/Partial Automation) to SAE Level 4 (High Automation) has come at a pace that has apparently taken even seasoned industry watchers by surprise.

A sharp increase in investment and acquisition activities over the past year has clearly helped accelerate the development progress across the industry as a whole. Global investments in autonomous vehicle (AV) technology skyrocketed in 2017 to more than $4 billion across 69 deals, up from $626 million across 34 deals in 2016. Tech giants like Google are pulling out all stops, expanding their self-driving teams and racking up millions of autonomous miles in real-world testing. All major automotive brands, meanwhile, are investing heavily in acquisitions and partnerships to wrangle for pole position in the impending disruption. The component ecosystem, which includes high-tech semiconductors, sensor systems, path planning algorithms, and many more, is also buzzing with alliances, acquisitions and VC investments.

All these factors are helping the industry to significantly and continuously accelerate the cycle time between R&D and production engineering.

Today, we are closer than ever to a driverless future and to the promise of improved road safety, lower traffic fatalities, reduced urban congestion and lower emissions.

As the promise slowly but surely morphs into reality, it becomes necessary to examine some societal implications of this groundbreaking shift. Take trucking for instance, a sector that is expected to lead the charge to a driverless future. If the political consequences of displaced jobs in the coal industry in the US are any indication, the future of truck drivers will be a key talking point in political narratives going forward. And given the fact that there are probably a thousand times as many truck drivers as there are coal miners in the US, the debate can only be expected to be more intense and contentious. Stakeholders in the AV ecosystem need to be thinking about that. The approach the industry takes to addressing issues like fleet driver unemployment will have a huge bearing on speed of adoption and growth trends in the industry.

However, there still is time to create a strategy that assures an inclusive future for AVs. Despite the hype that surrounds us, high automation already is a significant milestone: the end-of-this-decade ambitions announced by most players fall in the high automation rather than full automation category. And full automation capabilities, that completely eliminate the need for human intervention or even steering wheels and pedals, are still decades away due to numerous key challenges.

Road Safety

Let’s start with safety, especially since a 90-percent reduction in traffic fatalities has become a basic expectation of driverless technology. Today, 90 percent of fatal accidents are attributed to driver error, with mechanical and environmental factors accounting for the remainder. It is therefore arguably true that driverless technologies can increase net safety gains by eliminating the emotional, behavioral, or medical variables that cause driver errors.

We are still in the early phase of AVs and therefore pointing fingers on their flaws may be too harsh. Nevertheless, occasional discoveries of AVs’ weaknesses raise several significant questions regarding the robustness of autonomous systems to completely replace human input. For instance, Volvo Australia has shown that autonomous systems currently have some difficulty in recognizing and accurately detecting the location of kangaroos, a skill that comes naturally to human drivers. This is important, because there are over 16,000 collisions with kangaroos per year in Australia.

A few months ago, GM was sued by a motorcyclist in world’s first lawsuit to involve an AV, and just a few days ago, the world has seen the first fatal crash involving a pedestrian, which from the descriptions seem to be typical daily scenarios where the accident could have been avoided by a human driver using experience and intuition. Such instances call into question the ability of even fully autonomous vehicles to respond adequately to an exhaustive set of outlier scenarios, each of which is a potential accident.

This leads us to the question: what happens in the case of an unpredictable yet unavoidable accident.

Today, even the slightest public road scrapes involving AVs sets of an extensive debate about liability. The general view seems to place the onus of liability on manufacturers. But in the case of an accident involving two fully autonomous vehicles, will it be required to drill liability down to the level of mapping providers, nav software, or hardware? Even as the questions rage, Mobileye (acquired by Intel for $15.3 billion) has proposed a Responsibility Sensitivity Safety model to bring more certainty to the question of liability, alongside a call for a collaborative approach to building liability standards.

“Ideas and proposals for dealing with liability are constantly rolling in”, says Jae-Yong Lee, General Manager at Rewired. “The immediate challenge is to manage the coexistence of autonomous vehicles — with varying levels of automation — and current street dynamics with conventional vehicles and pedestrians. This will require a range of policy interventions to integrate driverless vehicles into conventional traffic systems. The vehicle-pedestrian interaction paradigm will shift too as AVs start to coexist with regular vehicles. There’s a big question on the impact on pedestrian safety that companies are starting to think about.”

Such innovations will play a crucial role in successfully rolling out AVs in daily life, but ultimately, the rate at which policy evolves will have a definite impact on the mainstreaming of driverless technology. The focus has to be on a wide range of factors, such as potentially assigning smart lanes for AVs, deploying intelligent traffic signaling systems and instituting a smart infrastructure strategy that eventually enables vehicle-to-everything (V2X) communications.

Manufacturing and Engineering Challenges

Cost is currently a huge challenge for the AV industry especially when it comes to building any meaningful scale in commercial applications. Just five years ago, lidars, a foundational mapping component in AVs, cost $80,000 a piece. Today, a similar sensor costs around $5,000 with sensor manufacturers working away on solid state solutions to break through the $1,000 barrier. And each AV comes fitted with a complex array of sonars, radars, lidars, cameras, etc, all of which can price the concept out of the market.

“Until recently, OEMs and Tier One suppliers weren’t very price-sensitive and were willing to stomach the high price of sensors as part of their R&D expenditures,” says Lee. “This has changed. Manufacturers are in a race to make AVs scalable and commercially viable. In parallel, assembling state-of-the-art components together also translates into other challenges, such as quality assurance or supply chain resilience. Already today, the industry’s push for more complex and simultaneously smaller sensors is making buyers wait up to six months for the sensors to be delivered.”

Then there is the challenge of balancing affordability with mission-critical performance. Concerns about the cost-capability imbalance of external solutions have driven Google’s Waymo to (reportedly) design and build all its self-driving sensors in-house. This approach should in theory allow Waymo to have tighter control over all software and hardware decisions and deliver a perfectly integrated system. But this is not an approach that is affordable, or practical, for every player in the AV industry.

Finally, there are the divergent approaches that different companies are taking to the shared dream of full autonomous vehicles. For instance, Tesla’s Elon Musk is confident of delivering level 5 autonomy without the need for any lidar. But this position has attracted some dismissals from an industry that is betting on these sensors being a necessary part of the formula for full automation. There is some research, however, supporting Musk’s position, but only in the context of parking lot speeds.

Software Integration

Fully integrated solutions will be required to power the industry to full and functional autonomy.

“It is going to take a combination of sensor types and sensor fusion models to make AVs a reality,” explains Gleb Chuvpilo, Venture Partner at Rewired. “Today, there are a range of hardware options competing against each other that will soon undergo some form of consolidation. The next big milestone for the industry will be represented by comprehensive solutions that can combine sensor data, make sense of it, and then deliver it to path planning and ADAS (advanced driver assistance systems) applications.”

Put in simple terms, a self-driving solution is an array of data-gathering sensors and devices on one side, constantly exchanging communications with an intelligent learning system on the other. The quality and efficiency of the interactions between these two systems will be critical in ensuring the accuracy and reliability of autonomous decision-making.

So the more data there is, in terms of both volume and variety, the larger the opportunity to create a more sophisticated and versatile self-learning ecosystem. In this context, Tesla is way ahead of the competition and in line to hit 11 billion miles of accumulated driving data by 2020.

But some companies, like FiveAI for example, are also turning to software defined innovations to get around this practical impediment of mapping road miles. Rather than waiting to “map 37.2 million kms of road across the planet”, FiveAI is focusing on high-performance AI to get to level 5 autonomy. A multi-sensor approach coupled with a robust sensor fusion solution and an intelligent path planning algorithm that does not have to pre-process millions of miles of data will be critical to reducing the time it will take to get to Level 5 automation.

Wrapping Up

The global AV industry today is poised at a point where the focus is more on “are we there yet?” than “how will we get there?”. Across the board, key technical and technological possibilities and challenges are being identified and addressed as we speak. But technological superiority alone cannot provide substantial or sustainable competitive differentiation. The true challenge will be to determine the best product-market fit and deliver capabilities that are aligned to consumer needs, while considering as well as anticipating regulatory hurdles, supply chain challenges, and people’s sentiments towards AVs on public roads.

SHARE

Author:

Posted on: March 23, 2018





Read more articles


Lidar Sensors: Applications and Opportunities

In March of 2004, fifteen self-driving cars raced across the Mojave desert in the inaugural Darpa Grand Challenge. Just over a decade later, autonomous vehicles are the “it” thing, thanks in no small part to Light Detection and Ranging (lidar) sensors.

Author: Rewired

Posted on: Oct 27 2017
SHARE

Launching Rewired, a Robotics-Focused Venture Studio and Fund

Santiago Tenorio is a Venture Partner at Rewired. Rewired is officially open for business. We have officially launched our new venture studio, focused on robotics and artificial intelligence. To start, we are investing $100 million in applied science and technologies that advance machine perception.

Author: Rewired

Posted on: July 26 2017
SHARE