Autonomous Driving Self-Supervised Learning

**Common mistakes:** Treating this as a purely technical deep-dive rather than a domain-knowledge and culture-fit conversation; Failing to connect personal labeling/annotation experience directly to AV-specific challenges; Not demonstrating genuine curiosity about Waymo's product roadmap or city expansion plans during the discussion **Interviewer hints:** HM proactively introduced team context and SSL landscape, effectively scaffolding the discussion rather than cold-questioning **What passers do:** Engaging interactively with the HM's domain introduction rather than passively listening; Showing crisp awareness of annotation vs. labeling distinction with AV-specific examples; Connecting prior batch-labeling or ML infra experience explicitly to Waymo's self-supervised learning pipeline needs **Why people fail:** Candidate was well-prepared technically but apparently did not sufficiently demonstrate product/domain alignment or proactive engagement during the HM's SSL discussion; Outcome: HM round not passed; all other Waymo roles blocked as a result **Edge cases probed:** Real-time model response under previously unseen adverse weather (out-of-distribution generalization); Distinction between 'annotation' and 'labeling' terminology in AV data pipelines **Alternative approaches:** Focus purely on supervised learning limitations (Highlights the bottleneck problem but misses the opportunity to discuss SSL mechanisms and how they apply to AV-specific data modalities (LiDAR, radar, cameras).)