Practical AI

Open Source Self-Driving with Comma AI

99 snips
Apr 16, 2026
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

Comma Keeps Driving Decisions Fully On Device

  • Comma keeps inference fully on-device; connectivity is only needed for updates and uploading data for training.
  • Harald Schäfer says the data center exists for training and simulation, not for user-facing runtime services.
INSIGHT

Heat Limits Edge Compute More Than Algorithms

  • Edge autonomy still faces harsh thermal limits: Comma's windshield device uses far less compute than Tesla's FSD hardware.
  • Harald Schäfer says an external GPU could unlock roughly 100x larger models and noticeable gains like better traffic-light recognition.
INSIGHT

Comma Targets Simple Retrofit And Familiar Controls

  • Installing Comma is designed as a simple retrofit: connect near the rearview mirror, stick the device to the windshield, and use the car's cruise-control button.
  • Harald Schäfer says the current product already feels best on highways, while city red-light behavior is still being polished.
Get the Snipd Podcast app to discover more snips from this episode
Get the app