Tech, ExplainedIndustrial tech and mobility
Product

Autonomous vehicle simulation

The rare crash, rehearsed a million times virtually.

BEFORE ITAFTER ITAutonomousvehicle simulationApplied IntuitionCOVERSVirtual roadsRare eventsTest scoringHD maps↗ GEOSPATIAL DATAsuppliesvirtual roadsSimulation software (CAE)↗ DIGITAL MANUFACTURINGprovidesphysicsmodelsreportstest resultsAutonomous driving softwarereportstest resultsAdvanced driver assistancesystems (ADAS)tests dronesoftwareDrone autonomy software↗ DRONE TECHNOLOGYCONCEPTS IT USESLevels of autonomyBEFORE ITAFTER ITAutonomousvehicle simulationApplied IntuitionCOVERSVirtual roadsRare eventsTest scoringHD maps↗ GEOSPATIAL DATAsuppliesvirtual roadsSimulationsoftware (CAE)↗ DIGITALMANUFACTURINGprovidesphysicsmodelsreportstest resultsAutonomousdriving softwarereportstest resultsAdvanced driverassistancesystems (ADAS)tests dronesoftwareDroneautonomy software↗ DRONE TECHNOLOGYCONCEPTS IT USESLevels of autonomyBEFORE ITHD maps↗ GEOSPATIAL DATA↓ supplies virtual roadsSimulation software (CAE)↗ DIGITAL MANUFACTURING↓ provides physics modelsAutonomous vehicle simulationApplied IntuitionVirtual roadsRare eventsTest scoringAFTER IT↓ reports test resultsAutonomous driving software↓ reports test resultsAdvanced driverassistance systems (ADAS)↓ tests drone softwareDrone autonomy software↗ DRONE TECHNOLOGYCONCEPTS IT USESLevels of autonomy
The simulator creates virtual roads and rare events, feeds them to the driving software as if from sensors, and scores how the software drives.
opens a page inside this one a page in the same scene a page in another scene a concept page a label only, with no page

Autonomous vehicle simulation is software that tests driving or flying software over millions of virtual kilometres, in a computer, instead of on real roads. It also covers drones and aircraft software, which have the same problem.

The problem is this. The dangerous moments that driving software must handle are rare. A child runs out between two parked cars. A truck loses its load on a highway. A driver comes the wrong way on a one-way road. A company cannot wait for these to happen, and cannot stage them safely on a real road. In a simulator, engineers can create them as many times as they like.

A simulator builds a virtual road, with virtual cars, scooters and people that move. It then gives the autonomous driving software the same kind of data that its sensors would give. The software drives the virtual car. The simulator records whether it stayed in lane, kept a safe distance and avoided a crash.

Imagine a team at an engineering centre in India that is testing the emergency braking of a car, for Indian highways. It wants to test one rare case: a cow steps onto the road at dusk. It creates the case once. Then it changes the cow's position, the car's speed and the light, to make 10,000 versions. It runs all of them in a few hours. The ones where the software fails show the engineers what to fix.

A simulator is different from a test on a closed track. A test track is real, so it is realistic, but each run costs time and money, and dangerous cases cannot be made safely. A simulator is cheap and safe, but it is only as good as its model of the world. If the virtual scooter does not move like a real one, the test proves little. So companies use both.

The simulator also tests driver assistance, and it can use the same lane detail as an HD map. It builds on methods from general engineering simulation software and from embedded system simulation. Drone software is tested the same way, as described under drone autonomy software. Carmakers and defence companies buy simulation to test the software they build. The leading makers of this software are based abroad, but Indian engineering centres use it daily.

Tech, Explained is a part of ingrid.fyi. Every page explains one piece of technology in plain words.