Embark SPAC Presentation Deck

Proprietary Active Learning System Drives Rapid Improvements Using Real World Situations To Rapidly Improve Performance 1000 Dahil "1 NORTHERN GENESIS II 1250 Edge Cases Irregular Objects pie 100 200 Irregular Conditions 100 800 1000 1750 what c Extremely ca Somewhat Uncertan ● ● Valuable Data Active Learning Model Training Selection Engine Deployed Model / Data Gathering Finding and solving edge cases is critical to the deployment of safe self-driving Our Active Learning pipeline is fed by uploading fleet data to our cloud-based selection engine The selection engine identifies edge cases by uniformly sampling and analyzing detection uncertainty for object existence, class and position across multiple permutations of the deployed model, culminating with a rank order of relative data value With the most valuable data automatically identified, we can focus our labeling and training efforts to provide the quickest, most effective feedback loop for the Embark Driver – resulting in constantly improving performance 11

View entire presentation