In a 2020 interview, Toyota Research Institute leaders Gill Pratt and Wolfram Burgard argued that self-driving depends on more than recognizing objects: a car also has to anticipate what people will do and choose a safe response. Their remarks explain why AI advances had not, in their view, settled the hardest parts of autonomous driving. They were discussing the state of the field at the time—not Toyota’s current vehicle lineup or a present-day forecast.
The interview was conducted by IEEE Spectrum senior editor Philip E. Ross at TRI’s Palo Alto offices and was condensed and edited for clarity. Its central idea is that a self-driving system must solve a connected chain of problems—and that strong performance at one link does not guarantee competence across the chain.
How did Pratt and Burgard break down the self-driving problem?
Gill Pratt described three stages: perception, prediction, and planning. In his words, “There are three different systems that you need in a self-driving car: It starts with perception, then goes to prediction, and then to planning.”
- Perception: Interpret sensor inputs to understand the surrounding scene.
- Prediction: Estimate how other road users and people may behave next.
- Planning: Choose the vehicle’s response based on the scene and likely future events.
Pratt singled out prediction as the most difficult stage: “The one that by far is the most problematic is prediction.” A system can identify a person, car, or obstacle and still face uncertainty about what that person or vehicle will do. That uncertainty then affects the action the car should take.
#1 Best Overall
- LAMBORGHINI REVUELTO SUPERCAR MODEL CAR KIT - Build the cutting-edge Lamborghini Revuelto, the brand’s first hybrid V12 supercar, recreated in highly detailed 1:24 scale for collectors and car enthusiasts.
- HIGHLY DETAILED 101-PIECE KIT - Includes multi-piece interior, detailed chassis, and precision-molded components that bring this next-generation supercar to life.
- REALISTIC V12 ENGINE DETAIL - Features a highly detailed hybrid V12 engine with opening engine cover, allowing you to showcase the heart of Lamborghini performance.
- POSEABLE & CUSTOM FEATURES - Front wheels can be positioned left or right, and the interior includes optional left- or right-hand drive configuration for added realism.
- AUTHENTIC INTERIOR DESIGN - Detailed cockpit with sport bucket seats and driver-focused layout, inspired by the real Revuelto’s aerospace-style interior.
What do cameras, lidar, and radar contribute?
Wolfram Burgard described cameras, lidar, and radar as complementary sources of information. Combining them can help a system build a picture of its surroundings, but it does not make interpretation automatic. Sensors can offer different viewpoints, and range estimates can be ambiguous; the system still has to reconcile what the inputs mean.
The interview’s point is not that any one sensor is sufficient or that combining sensors eliminates uncertainty. Sensor fusion supplies evidence for the perception task; prediction and planning remain separate challenges.
Rank #2
- Revell Model Kit #14531, Skill Level 5, Contains 77-Parts, Recommended for ages 13 and up
- Detailed 440+6 V-8 engine, hood scoop that can be positioned open or closed
- Detailed interior with separate side panels and a front bench seat
- Molded in white and clear with chrome plated parts and black vinyl tires
- Illustrated assembly instructions. Requires paint and glue. (not included)
Why did Pratt say deep learning has limitations?
Pratt characterized deep learning as “high-performance pattern matching.” He regarded pattern recognition as powerful, but not equivalent to all the reasoning required to drive: detecting a familiar pattern does not by itself establish how another person will act or what response is safest in an unfamiliar situation.
The interview therefore favored combining structured components and different methods over assuming that a single learned, end-to-end mapping would solve the entire driving task. Burgard framed the uncertainty about what comes next this way: “We are now in the age of deep learning, and we don’t know what will come after.” This was a critique of relying on pattern recognition alone—not a claim that autonomous driving is impossible or that progress had stopped. Pratt was explicit: “There isn’t anything that’s telling us that it can’t be done; I should be very clear on that.”
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Rank #3
- Durable Metal Racing Car Set - Our building toys car set includes 287 parts and pieces such as nuts, bolts, small screwdrivers and wrenches. The model car toys takes a little time and patience which is an immersive build for kids who love challenges. Building toys for boys age 8-12. STEM education through playing fun.
- Detailed Assembly Instructions- Each step of the installation has a detailed instruction manual, it makes each step clear and controlled, children can effortlessly complete the model assembly.
- STEM Educational Building Toys - This stem kits for kids age 8-12 offers a great hands-on experience. It helps kids develop spatial thinking, hands-on skills, hand-eye coordination, and teamwork abilities. Really suitable for model car fans and DIY enthusiasts and erector sets for adults.
- High Quality and Safe Materials - Made with high-quality metal components, this model truck ensures strong structural stability after assembly without loosening. The finished design includes smooth-rolling wheels and adjustable parts for added play value. Ideal for stem activities for kids aged 8-15.
- Gift for Kids - This model racing car kit for kids 6 7 8 9 10 11 12 year old boys girls. It makes an excellent gift for birthdays, Christmas, holidays. It ignites children's curiosity and provides an educational and engaging hands-on activity.
How can an operational design domain make driving more manageable?
An operational design domain (ODD) specifies the conditions in which a system is intended to operate. In the interview, Pratt described boundaries that can include geography, weather, traffic, and speed. Limiting those conditions narrows the situations the system must handle; it does not eliminate difficult cases within the boundary or make performance outside it established.
Pratt’s discussion of lower-speed urban driving and highway use as potentially more tractable settings was his assessment in 2020, not a current forecast. He also pointed to weather and unexpected debris as examples of complications. The useful question when evaluating an autonomy claim is therefore not simply whether a car is “self-driving,” but which tasks it performs and under what conditions.
Rank #4
- A PERENNIAL CLASSIC FOR EXPERIENCED MODELERS: AMT’s 1/25 scale 1967 Shelby GT-350 is the perfect project for muscle car builders. Add it to your collection today!
- FEATURE PACKED: The 1967 Shelby GT-350 is a historical muscle car that stands alone! Fully paintable, it features authentically detailed water-slide decals. Build it up right with a Ford 289 V8 engine and all the Shelby touches—custom rear lower pan, chrome wheels, custom taillights and front grille, bucket seats, and more. Wrapped up nicely in retro-styled packaging!
- QUICK SPECS: 1/25 Scale, 123 Parts, 7 inches long once assembled. Parts molded in white, clear and transparent red, with black vinyl tires. Some parts are chrome plated. Plastic model kit paint and cement required (Testors or Tamiya brands recommended). Paint guide included in kit. Skill Level 2. For ages 10+
- THE PERFECT PRESENT: Don't know what to get dad for his birthday? Or maybe you have an avid hobbyist or collector in your life. This model kit makes an ideal gift for any occasion!
What did Toyota mean by Guardian and Chauffeur?
The interview used two names for distinct concepts. Guardian was described as assistance that backs up a human driver; Chauffeur was described as a more futuristic system intended to replace the driver. The article also identified a Platform 4 test vehicle based on the Lexus LS in connection with Level 4 Chauffeur development.
These are descriptions from the interview, not evidence that either concept or the test vehicle is currently available to consumers. They also illustrate why “autonomy” alone is too broad to compare systems: a driver-support system and a driver-replacement system assign responsibility differently.
Best Value
- 192 PCS METAL RACING CAR KIT FOR 8+ AGE: Kids can build a realistic racing car using real screws, nuts, bolts and tools for an exciting hands-on engineering experience. A fun car building kit for boys 8-12 that encourages creativity, problem-solving and STEM learning through interactive play
- REAL TOOLS AND EASY BUILD: Includes illustrated instructions, screwdriver and wrench designed for beginner builders. This model car kit helps kids confidently assemble their own racing car while improving patience, focus and logical thinking skills
- REALISTIC MECHANICAL DESIGN: Features durable metal parts, smooth-rolling wheels and realistic racing car details that create an authentic building experience. A fun STEM engineering toy and erector set for kids who enjoy construction toys and model kits
- STEM LEARNING THROUGH PLAY: Designed to improve fine motor skills, creativity, spatial awareness and engineering thinking through hands-on building activities. Suitable STEM toy for ages 8-13, homeschool projects and screen-free educational fun
- EXCITING GIFT FOR AGES 8-13: An exciting birthday, Christmas and holiday gift for boys, girls and teens who love model car kits and engineering toys. Encourages family bonding and hours of creative building fun
What safety standard should an automated car meet?
The interview raised a question beyond engineering: what level of performance should people accept, and would the public react differently to crashes involving automated cars? The comparison depends partly on the benchmark—human drivers alone, or human drivers supported by active safety systems—and on what the automated system is actually responsible for doing.
Pratt speculated that people may respond differently to a crash involving automation, including because of empathy for the person in the vehicle. He labeled that point as speculation; it should not be treated as a measured public-opinion finding. The interview does not establish a numerical safety threshold or provide current comparative performance data.
How to read the interview today
Pratt and Burgard’s comments are most useful as a framework for understanding the problems autonomous vehicles must address: perception, prediction, and planning; the limits of pattern recognition; the conditions defined by an ODD; and the distinction between driver assistance and driver replacement. Because the interview is historical, it should not be read as a description of Toyota’s present products or as an update on the current state of autonomous driving.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
