Self-Driving Cars Need to Be More Reliable

Reliability is the underlying issue with all the challenges still facing self-driving cars. While we have seen great strides in moving towards a more autonomous and smart vehicle approach, reliability in cluttered environments and inclement weather remains an elusive goal.


Cars have been driving autonomously for longer than most people think. In 1995, researchers from CMU drove a car semi-autonomously across the country with less than 3 percent human interaction. At the Idaho National Lab, Humvees and Jeeps were driving fully autonomously to do radiation surveys and environmental monitoring over a decade ago.

Now autonomous driving is in the news every day. And with all the hype, people often ask me what the holdup is. Is there anything left that needs to be done? Of course there’s still work to be done. Programming the rules of the road, providing required infrastructure, and addressing the limitations of GPS positioning are remain priorities.

The underlying issue with all these challenges is reliability. While we have seen great strides in moving towards a more autonomous and smart vehicle approach, reliability in cluttered environments, urban canyons and inclement weather remains an elusive goal.

Rules of the Road and Required Infrastructure

The challenge of interpreting all the signs and interaction cues associated with daily driving is generally called rules of the road. Thousands of tragic accidents each year are preventable if we can get the technology right, but researchers are still struggling to get cost-effective solutions for tough environments such as pedestrian-rich environments, areas with construction, and inner cities. Although most researchers don’t want to admit it, we may have to help the vehicles out by changing our infrastructure.

Must-Read: Guidelines for Self-Driving Cars: Everything You Need to Know

This should not come as a surprise to us since we had to add billions of dollars worth of infrastructure to our roads to create a human-readable world. We sometimes expect intelligent robots to just “figure it all out,” but we already know that intelligent humans require plenty of help to navigate our roads.  This is why we have embedded countless signs, signals and lights into the environment. If you removed all the signs from the roads and asked me to navigate across the country to Las Vegas, I might eventually make it, but not in an efficient fashion.

Robots also need help,  and infrastructure is the best way to ensure their success. Infrastructure tends to be viewed at first as a negative, but whether laying down rails, putting up traffic lights, paving an interstate highway, or stringing telephone cables across the nation, the success of new technology is directly dependent on proper infrastructure.

Limitations of GPS Positioning

Think about the impact GPS has already had on how we communicate and move in our daily lives. Now imagine if that position information was more accurate, worked indoors and included accurate directional heading as well as position. We have come to accept the limitations of GPS, but our increasing emphasis on autonomy, efficiency and situational awareness is prompting us to reconsider current limitations. GPS positioning works well within a telecommunications context where 99 percent reliability is permissible. If your cell phone tells you to make a left hand turn off the Bay Bridge, you just ignore it and shake your head. In contrast, applications we are currently contemplating, such as self-driving cars and automated drone delivery, are fundamentally different. If one out of one hundred cars veers out of their lane unexpectedly or if one of a hundred forklifts drive off the edge of the loading dock, the results could be disastrous.

Blind dependence on GPS can limit efficiency and reliability. GPS can’t direct a robot through a narrow doorway or prevent a collision inside a parking garage. It can’t slide your car into a parking space or allow a convoy of trucks to keep a meter apart. We have tried to solve this problem by adding sensors onto vehicles that build up maps and localize, but we can’t put these on most of the things we’d like to track and control because the sensors are either too big, too power hungry or don’t work in dirty, dusty, dynamic environments.

The opportunities are immense if we can improve the technology, but to do so we may have to think outside the box – enter peer-to-peer positioning.

Peer-to-peer Positioning

Ants don’t have maps and don’t know their global position, but they can build whole civilizations and navigate using pheromones they place in the environment. These pheromones serve as embedded signposts that direct both individual and group behavior while serving as an implicit framework of peer-to-peer positioning. What if we built a similar peer-to-peer system to underpin our current focus on global, centralized control?

In a peer-to-peer system, behavior stems from the local environment, making it easier to adapt to a particular environment and situation. Distributed systems are more fault tolerant because there is no single, centralized point of failure. It’s not just an engineering issue. Do we want to depend on our neighbors or on a central bureaucracy? Enhanced awareness of our local environment may help us look at the world differently. Centimeter level, real-time position tracking can help us answer questions of why we do things and help us form connections in space, time and purpose that are not possible today.

The plug-and-play 5D Universal Position Node being used this month on a light post to provide accurate, reliable positioning for ten city blocks. (Credit: G&G LED Lighting)

As engineers and decision-makers contemplate a move away from the global position paradigm, it may leave them with a strange feeling of vertigo. Like pheromones in an ant colony, our world will fill with digital signposts or “tags” that create a robot-readable world. Some tags will be anchored into the world as infrastructure, built into light bulbs, traffic lights, and signs. These infrastructure tags will enhance or simply replace GPS, feeding into existing software applications that already expect GPS coordinates. In this tag-enabled universe peer-to-peer positioning can guide both humans and robots, creating a common framework for shared understanding and collaboration.

What’s the Problem with Cameras and LIDAR?

It may seem that our transportation system would be the most obvious place to apply peer-to-peer positioning. After all, even if GPS did not have problems with accuracy and reliability, common sense tells us that it is just plain silly for cars 10 feet apart to depend on satellites 20 kilometers away to position themselves.  Nonetheless, without a way to mandate the adoption of peer-to-peer technology across the board, many companies are betting on classical artificial intelligence (AI).

AI techniques have long promised to use sensors such as vision and lasers to make cars smarter and less dependent on GPS by allowing the car to decide its position in a map. These sensors can provide benefit but our extensive work with the military indicates that these sensors struggle with rain, dust, snow, fog and, as a recent Tesla crash made clear, even bright sunlight.

Must-Read: What is Lidar and How Does it Help Robots See?

The same reasons these optical sensors are challenging for the military also make them problematic for our roads. Even without environmental challenges, these sensors will always be limited to line of sight. None of the traditional range sensors (e.g. stereo vision, infrared, sonar, radar, LIDAR) have the ability to see through cars or around corners. Also, if the world or environment changes since the time the map was made, the robot may be confused. Even if we ignore all of these problems, we are still left with the fact that the sensor suite on most self-driving cars totals over $300,000—a very costly solution.

The Future of Self-Driving Vehicles

For robotic applications, the most important metric will be reliability. There is no better way to do this than by standardizing a system of peer-to-peer positioning and connectivity that can help both human and AI driving systems behave appropriately together. The Department of Transportation (DOT) is pursuing a connected vehicle solution that will usher in the next generation of intelligent transportation.

On August 22 of this year, the DOT released results from the New York City connected vehicle test-bed that indicated accurate, reliable positioning remains a concern near tall buildings, under bridges and in tunnels. The DOT is currently looking for alternative positioning technologies to solve this problem. Formal acknowledgement that there are limits to the efficacy of GPS opens up possibilities for new and better technologies to be introduced.

In the future, mesh networks of reliable relative ranging can allow diverse vehicles to use a shared backbone of safe, coordinated motion. The cars can communicate directly to one another in a swarm-based fashion (i.e. apart from any infrastructure), but they can also use roadside tags in light posts and other forms of infrastructure to know precisely where they are in global coordinates. This makes the autonomy problem much easier for individual vehicles and also makes it possible to orchestrate multiple vehicles as a team.

In addition to supporting cars, this enables many new forms of autonomous shared mobility. Single-person pods, low-speed electric vehicles, autonomous cars, human-driven cars and autonomous ride -sharing systems can virtually snap together despite their different manufacturers and control systems. Working together to anticipate traffic light changes, these vehicles can smoothly accelerate in unison and perform predictive braking when necessary. Imagine a pod that comes to meet you right at the gate as you disembark from the airplane. It carries you through the airport and merges with another pod that has your luggage before dropping you off directly at the train station or curbside for a car pickup.




About the Author

David Bruemmer · David Bruemmer provides leadership and vision as CTO of 5D. He enjoys finding ways to fuse emerging science and engineering into innovative technologies that can change the way robots interact with humans and their environment. Bruemmer has developed and patented robotic technologies for landmine detection, defeat of improvised explosive devices, urban search and rescue, decontamination, chemical plume tracing and decommissioning of radioactive environments, air and ground teaming reconnaissance systems, facility security and a variety of autonomous mapping solutions. Bruemmer has authored over 60 peer reviewed journal articles, book chapters and conference papers in the area of intelligent robotics. He is a winner of the R&D 100 Award and the Stoel Reeves Innovation Award.

Before co-founding 5D, Bruemmer provided technical, managerial, and programmatic leadership at the Idaho National Lab for a diverse, multi-million dollar research and development portfolio. His team included a multi-disciplinary group of scientists and engineers including mechanical, electrical, computer, and human factors engineers; cognitive psychologists; computer scientists; and roboticists. Bruemmer established strategic research collaborations with many universities, government labs, and industry partners. Between 1999 and 2000, Bruemmer served as a consultant to the Defense Advanced Research Projects Agency (DARPA), where he worked to coordinate development of autonomous robotics technologies across several offices and programs.
Contact David Bruemmer: monica@5Drobotics.com  ·  View More by David Bruemmer.




Comments

ConnectomicAI · September 23, 2016 · 6:15 pm

David: You hit on a subject near and dear to my efforts. I agree with your premise and I think your ideas regarding connected vehicles is a very good idea assuming hackers and cars in need of repair do not mess up the abilities of your car. However, there are AI techniques like I am developing that surpass the Deep Learning paradigm which I think we can both agree is very limited and will not progress any further that what we are already seeing. I work with artificial biological intelligence and specifically with Connectomic AI which allows the vehicle to truly operate as if a human were operating the vehicle. The thought might be that humans are not reliable so why would this be better? If humans knew the rules of the road 100% and were capable of concentrating on their driving 100% of the time, never being distracted or day dreaming, I feel we would have a huge reduction of traffic incidents to nearly zero. The problems we face today are drivers do not know the rules and more often make up their own rules and of course, no one can concentrate 100% of their time on their driving but a “robot” outfitted with the right intelligence could. Furthermore, getting to Las Vegas could be a dangerous proposition for today’s self driving cars because they rely on technology that most likely would not be available on some stretches of highway from SF to LV. Connectomic AI works like I do and does not need GPS or to talk to other vehicles to get from point A to B.


ConnectomicAI · September 23, 2016 at 6:15 pm

David: You hit on a subject near and dear to my efforts. I agree with your premise and I think your ideas regarding connected vehicles is a very good idea assuming hackers and cars in need of repair do not mess up the abilities of your car. However, there are AI techniques like I am developing that surpass the Deep Learning paradigm which I think we can both agree is very limited and will not progress any further that what we are already seeing. I work with artificial biological intelligence and specifically with Connectomic AI which allows the vehicle to truly operate as if a human were operating the vehicle. The thought might be that humans are not reliable so why would this be better? If humans knew the rules of the road 100% and were capable of concentrating on their driving 100% of the time, never being distracted or day dreaming, I feel we would have a huge reduction of traffic incidents to nearly zero. The problems we face today are drivers do not know the rules and more often make up their own rules and of course, no one can concentrate 100% of their time on their driving but a “robot” outfitted with the right intelligence could. Furthermore, getting to Las Vegas could be a dangerous proposition for today’s self driving cars because they rely on technology that most likely would not be available on some stretches of highway from SF to LV. Connectomic AI works like I do and does not need GPS or to talk to other vehicles to get from point A to B.


Log in to leave a Comment


in the Future Tech Hub

Editors’ Picks

CES 2018 AI Conference Schedule
Robotics Trends' AI conference at CES 2018 examines recent developments, current applications, and...

Jibo Music Brings iHeartRadio to Social Robot
ibo and iHeartRadio have teamed up to launch Jibo Music that will...

Japanese Startup GROOVE X Goes Viral as Teaser for LOVOT Robot
GROOVE X is teasing its LOVOT companion robots that are scheduled to...

What Humanoid Backflips Mean for Robot Agility
In just 24 months, machine agility has gone from the Keystone Kops, to...