Perception and localization

Leverage simulation, re-simulation, and synthetic data to inform autonomous vehicle (AV) decision-making.

“We use the most powerful and flexible simulation platforms in the self-driving industry.”

Mike Carter
Founding Engineer at Kodiak

Customer challenge

Testing AV perception and localization systems in the real world is time-consuming and often difficult to execute and repeat at scale.

Real-world operations don’t capture all situations

One situation has many edge cases and variations

Certain events are dangerous to test in the real world

Collecting and labeling real datasets is costly

Applied’s solution

Applied’s solution allows teams to test AV perception and localization systems at scale, identify edge cases, and ensure high-quality code.

Re-simulate failures found in the field

Generate synthetic datasets to train perception models

Leverage sensor simulation to iterate rapidly

Workflow

Develop high-performing perception and localization systems

01

Identify issues found in field testing

Search real-world logs to identify segments that challenged your AV stack, including issues like driver interventions.

Strada

02

Perform root cause analysis

Re-simulate the identified real-world logs to understand your AV stack’s performance and determine necessary adjustments.

Strada

Logstream

03

Improve and test perception modules

Modify perception modules or re-train machine learning (ML) models with synthetic data to address the identified issues. Iteratively run sensor simulations to monitor progress across purpose-built test cases, and introduce variations across behaviors, weather, and lighting to ensure robustness.

Spectral

Synthetic Datasets

04