Back for More -- TRI Chief Technology Officer James Kuffner inspired a lot of questions with his talk at NABU last month. Below, he answers the ones he couldn't get to at the meeting.
At the North American Business Update meeting earlier this month, one thing became clear: You guys have a lot of questions about the Toyota Research Institute.
TRI Chief Technology Officer took questions at NABU, , but time restraints left a lot unanswered. Luckily, Kuffner loves to talk about his business, so we sent him your questions we didn’t get to ask, and he took the time to answer them.
In some cases, we condensed or combined answers for length and brevity.
Team Member Question: Where are we compared to, say, Tesla? And why don’t we find ways to collaborate with them instead of competing with them?
We don’t make comparisons to others in this space, other than to say Toyota is one of the leading companies. Our goal is not to be first, but to ensure the technology works correctly before it’s deployed. We have had collaborations in the past, including a prior relationship with Tesla, and we continue to explore opportunities with companies pursuing like-minded goals. We have also made a few investments through our venture capital subsidiary called Toyota AI Ventures.
Who does TRI see as its main competitors for autonomous vehicles?
Nearly all the major automakers and a few suppliers have investments in developing automated technology. There are also several big-name Silicon Valley companies and many startups pursuing opportunities in this space.
Is TRI looking for new ways to make cars? For example, can we use Guardian technology for tow motors or fork lifts?
TRI’s role is to conduct advanced research in automated driving technology and present discoveries to Toyota for potential applications into the family of products. Our aim is to have both Guardian and Chauffeur autonomy software available for use in as wide a range of Toyota products as possible.
When will we see an autonomous test on public roads?
TRI has been testing autonomy carefully on public roads since the very beginning. We are currently testing autonomously on public roads in both Michigan and California. In other areas, we are conducting tests primarily aimed at data collection.
Does our autonomous vehicle technology detect approaching vehicles? Also, where does vehicle security stand while adding all the technology on the autonomous vehicle?
Our test vehicles have an advanced suite of sensors -- LIDAR, radar, cameras -- to detect objects in the environment and continuously estimate their positions, velocities, and predicted future trajectories. Our new test units can sense objects up to 200 meters in the complete 360-degree perimeter of the vehicle. Toyota Connected is working on ensuring the security of the network connection for the vehicle, which will also secure the internal autonomy functions.
How will we “train” autonomous vehicles to react in situations where a child or dog runs out on the street and evasive action is potentially dangerous to the occupant of the vehicle?
TRI uses artificial intelligence and machine learning in its automated driving development to predict what may happen in the vehicle’s environment and make decisions on how the vehicle will maneuver in response. One of the challenges will be to accumulate enough training data such that the system will learn the appropriate tradeoffs over time that meet or exceed human judgments.
How do TRI and Toyota Connected align with Toyota Motor Corporation (TMC) and TMNA R&D programs?
TRI and Toyota Connected work hand-in-hand with TMC and TMNA. Most recently, TRI tapped the expertise of TMNA R&D and Calty Design Research in the development of its new automated vehicle test platform 3.0, which was unveiled at CES (Consumer Electronics Show) in January. Additional collaborations are already underway on future vehicle prototypes.
How will our autonomous vehicles handle inclement weather, like snow and ice?
There is still a tremendous amount of research and development needed for automated vehicles to operate in all weather conditions. This is a challenge for everyone in the industry, which is one of the biggest barriers to overcome before achieving SAE (Society of Automotive Engineers) Level 5, which is defined as fully automated driving under any condition. This is also the reason that automated driving technology will likely appear commercially in a Mobility as a Service (MaaS) context with traffic and weather constraints (SAE Level 4) before maturing enough to be deployed at scale on personally owned vehicles.
How will our autonomous vehicle technology react in the case of a carjacking?
There is no blanket response to a scenario. There are an infinite number of low-probability, high-risk situations, often referred to as “corner cases,” that the artificial intelligence will need to handle in the future. We believe that appropriate responses to specific situations can be learned over time as the systems are deployed and data can be collected for training. But the carjacking example illustrates how difficult attempting to manually program an appropriate response to each individual situation can be.
Does the blockchain or distributed ledger technology play into the different technology we saw today, such as autonomous driving and connected vehicles or even robotics?
TRI is exploring use of blockchain technology, which we announced last year.
Can TRI provide equipment to Toyota team members to capture the necessary data?
TRI is currently focused on capturing sets of high-resolution data requiring a very sophisticated set of LIDAR (Light Detection and Ranging), radar, cameras and computer processing equipment, which can only be equipped on dedicated test vehicles. However, we are exploring lower-resolution and lower-cost data collection possibilities for the future.
How will autonomous vehicles handle state-to-state and country-to-country road laws? For example, a right turn on red, or creeping into an intersection for an unprotected left?
In a world that will be inhabited by both human-driven and AI-driven cars for the foreseeable future, it makes sense for the AI-driven car to “learn” the expected behaviors through data observations of human driving. In addition, the vehicles must obey regional and national laws. What is challenging is that there are often unwritten rules and expected behaviors that vary from state to state and country to country. Fortunately, machine learning systems can be configured to automatically learn these regional differences and unwritten rules of behavior and be trained to mimic them.
What’s the backup for when Chauffeur mode fails?
In addition to protecting the human driver, the Guardian system could also protect the Chauffeur system. This is an approach that we are exploring.
Will our autonomous cars be as pricey as, say, the Tesla Model S?
We are still in the early stages of development, so it’s premature to talk about a pricing strategy. The cost of deploying this technology is currently largely dependent upon the cost of the sensors and computing components. We expect that, over time, these components will become commoditized and will therefore enable the technology to be deployed across a wide-range of vehicle models and prices, much like how TSS (Toyota Safety Sense) has expanded to almost all Toyota vehicles.
What is TRI doing to ensure robots and cars don’t get too smart for their own good?
This is definitely something that TRI is attuned to, and there are actions that can be taken, like embedding safety buffers to override a system.
Will Toyota collaborate with other manufacturers so that, in the future, autonomous vehicles act together on the road instead of reacting to each other like drivers to today?
Toyota has been working on initiatives around vehicle-to-vehicle communications with government and industry partners for several years. We hope that it can be a reality someday.
When will Toyota team members in Ann Arbor be able to ride TRI autonomous shuttles to commute between buildings?
We’re still a few years away from the technology being ready for deployment.
What are TRI’s latest victories?
We just introduced our next-generation automated vehicle testing platform, Platform 3.0
, at CES in January. This vehicle has a very sensor-rich package that makes it one of the most perceptive automated driving test cars on the road. The sensing equipment has been blended into the Lexus LS vehicle design with a distinct appearance that is sleek and elegant.
By James Kuffner and Toyota team members