Foundations
Self-Driving Glossary
A plain-language glossary of the terms engineers, regulators and reporters use about self-driving cars, from ADAS and ODD to ego vehicle and disengagement. Each entry is a short, self-contained definition, consistent with the SAE J3016 standard where it defines the term, with a link to the lesson that goes deeper.
ADAS (advanced driver assistance systems)
Features that help a human driver, such as adaptive cruise control, lane centering or automatic braking. The human is still driving and must supervise at all times. In SAE terms these are Level 0 to Level 2 features.
AEB (automatic emergency braking)
A system that detects an imminent collision with a vehicle or pedestrian and brakes on its own if the driver does not respond in time. It has been required on all new cars in the EU since July 2024.
Automated driving system (ADS)
The SAE J3016 term for the hardware and software that together can perform the entire driving task on a sustained basis. It applies only to Levels 3, 4 and 5.
Autonomous vehicle
The everyday term for a vehicle that drives itself. SAE J3016 discourages it as ambiguous, since these vehicles still depend on outside communication and support, and prefers naming the SAE level.
Bird's-eye view (BEV)
A top-down, map-like representation of the scene around the vehicle. Models build it by projecting camera, lidar or radar data onto one ground-plane grid, which makes fusion and planning easier.
DDT fallback
The SAE J3016 term for the response when a system fails or the vehicle is leaving its operational design domain: either resume driving or reach a minimal risk condition. At Level 3 a human does this; at Levels 4 and 5 the system does.
Disengagement
When the automated mode switches off, because the technology fails or because a human must take over for safety. California requires companies testing autonomous vehicles to report them. It is a rough signal, not a direct safety measure.
Drive-by-wire
Steering, braking and throttle controlled by electronic signals to actuators rather than by mechanical or hydraulic linkages alone. It lets a computer command the vehicle directly.
Dynamic driving task (DDT)
The SAE J3016 term for all the real-time functions needed to drive in traffic: steering, speed control, and watching for and responding to objects and events. It excludes strategic choices such as planning the trip.
Edge case and the long tail
Rare, unusual situations, such as strange debris, unusual road users or odd weather combinations. The long tail means each one is rare, but together they are a large share of real-world risk and are hard to cover in training data.
Ego vehicle
The vehicle the system is reasoning from: our car, as opposed to the other road users around it. Engineers speak of the ego lane or ego speed in the same sense.
End-to-end driving
An approach in which one learned model maps raw sensor input directly to a driving plan or controls, instead of chaining separate perception, prediction and planning modules.
Geofence
A virtual geographic boundary that limits where an automated vehicle may operate. It is one part of the operational design domain, and robotaxi permits typically define it as the approved service area.
HD map
A high-definition map with centimeter-level detail on lane geometry, boundaries, traffic signals and signs. Vehicles use it as prior knowledge for localization, prediction and planning.
Imitation learning
Training a driving system to copy recorded human driving. On its own it struggles with situations the recorded driving never shows, so it is often combined with other training methods.
Lidar
Light detection and ranging: a sensor that measures distance with laser light and builds a precise 3D picture of its surroundings.
Localization
Working out exactly where the vehicle is and which way it faces, often to within centimeters, usually by matching sensor data to a map and combining satellite, inertial and wheel measurements.
Minimal risk condition (MRC)
The SAE J3016 term for a stable, stopped state that the vehicle is brought to after a failure or when leaving its operating conditions, to reduce the risk of a crash. Pulling over to the shoulder is an example.
Occupancy grid and occupancy network
An occupancy grid divides the space around the vehicle into cells, each marked free, occupied or unknown. An occupancy network is a neural network that predicts this grid directly from sensor data.
Operational design domain (ODD)
The SAE J3016 term for the specific conditions a driving automation feature is designed to work in: geography, road types, speed, weather, time of day and traffic. Outside its ODD, the feature is not meant to operate.
Perception
The part of the system that turns sensor data into an understanding of the scene: detecting, classifying and tracking vehicles, pedestrians, lanes, signs and free space.
Planning (motion planning)
Choosing what the vehicle will do next and computing a safe, comfortable and lawful path for it. A controller then carries out that path through steering, throttle and brakes.
Point cloud
A set of 3D points, each with coordinates and often an intensity value, typically produced by lidar. It is a raw geometric snapshot of the surfaces around the vehicle.
Prediction
Forecasting what other road users will do over the next few seconds, usually as several possible paths with probabilities, so the planner can react in advance.
Radar
A sensor that uses radio waves to measure an object's range, angle and speed directly. It works well in rain, fog and darkness but sees less detail than lidar.
Remote assistance
When a remote human gives advice to a driverless vehicle that is unsure what to do, such as confirming a path. The vehicle keeps doing all of the driving. It is not the same as teleoperation.
Robotaxi
A ride-hail vehicle driven by a Level 4 automated driving system, usually with no human driver aboard, carrying paying passengers inside an approved service area.
SAE levels (0 to 5)
The six levels of driving automation defined in SAE J3016. At Levels 0 to 2 a human drives and supervises. At Level 3 the system drives but a human must take over when asked. At Level 4 the system handles everything within its ODD, and at Level 5, everywhere.
Safety driver
A trained person in a test vehicle who monitors the automated system and takes manual control when needed.
Sensor fusion
Combining data from several sensors, such as cameras, lidar and radar, so each covers the others' weaknesses, giving a more accurate and robust view than any single sensor.
Simulation and scenario-based testing
Testing driving software in virtual worlds or on replayed recordings, often organized as defined scenarios such as a car cutting in. It covers far more situations, especially dangerous ones, than road testing alone can.
SLAM
Simultaneous localization and mapping: building a map of an unknown environment while tracking the vehicle's position within it at the same time.
Teleoperation (remote driving)
When a remote human performs some or all of the driving in real time, steering and braking over a network link. SAE J3016 calls this remote driving. It is not the same as remote assistance, where the vehicle still drives itself.
V2X (vehicle-to-everything)
Wireless communication between a vehicle and other vehicles, roadside infrastructure, pedestrians and networks, sharing position, speed and hazard information beyond what onboard sensors can see.
Vision-language-action model (VLA)
A large AI model that takes camera images and language as input and outputs driving actions, and can explain its reasoning in words. It is a newer branch of end-to-end driving.
Watch
Frequently asked
- What does ego mean in self-driving?
- Ego vehicle means the vehicle doing the sensing and deciding: the car itself. Everything else is described relative to it, as in ego lane or ego speed.
- What is an ODD?
- An operational design domain is the set of conditions a driving automation feature is designed to work in, such as specific roads, speeds, weather and times of day. A robotaxi's service area and weather limits are part of its ODD.
- What is a disengagement?
- When the automated driving mode switches off, either because the system fails or because a human must take over for safety. California requires companies testing autonomous vehicles to report them.
- What is the difference between remote assistance and teleoperation?
- With remote assistance, a remote human gives advice to a driverless vehicle and the vehicle keeps doing all of the driving. With teleoperation, also called remote driving, a remote human steers and brakes the vehicle in real time over a network link.
Keep learning, once a week
The weekly digest: the stories worth your time from the autonomous vehicle industry, hand-picked by people who work in it. No marketing gloss.