Foretify Evaluate

Data Curation

Unify, aggregate and cleanse data from both real-world and virtual drive logs to optimize the training and validation data set and identify gaps and redundancies
Search for similar or rare scenarios, to increase scenario diversity and relevance during AI model training
Prioritize high-value samples to improve AI model robustness and training effectiveness

Data Curation

ODD Coverage

Unify data from both real-world and virtual drive logs to measure the ODD coverage and identify specific gaps during validation
Improve productivity by focusing development, training and testing resources on high priority performance, quality and safety gaps

ODD Coverage

Safety Evaluation

Measure testing completeness and assess performance and safety to contribute to the safety case
Provide evidence for the safety case, for internal and external stakeholders

Safety Evaluation

Performance & Quality

Expose high-priority bugs (edge cases and unknown unknowns) that would have taken years to encounter in the real-world
Focus on resolving critical issues, propelling the data flywheel for faster development cycles and improved safety readiness at launch

Performance & Quality

Capabilities

Unified coverage view

Visualize progress against a high-level coverage plan driven by aggregating physical and virtual logs.

Single run debugger

Inspect key events and KPIs in the scenario context, focusing the debugging tools for improved efficiency.

Formal definition of abstract scenarios

Identify scenarios using a formal scenario language (OpenSCENARIO DSL) to ensure consistency and traceability across workflows.

Sensor Data Curation

Identify scenarios based on visual characteristics using natural language, complementing formal scenario search techniques. Foretellix provide an integrated solution built on NVIDIA Cosmos World Foundation Models.

Evaluation Libraries

Apply a library of scenarios, KPIs and coverage definitions to automatically detect and classify the ODD coverage and performance from the drive logs.

Triage

Apply automation to identify and analyze anomalies and critical issues. Compare different AV stack versions, to identify degradations.
Generate performance dashboards for transparent, data-driven views to all stakeholders.