Protocol

Research toward a safer autonomous world

A research lab working on perception safety for machines that will share space with people

01
asymmetry

Autonomous machines are entering everyday life faster than the frameworks meant to govern them

Humanoids are leaving controlled environments. Vehicles are negotiating with pedestrians. The governance scaffolding was written for an era when machines moved on rails.

02
perception

Every downstream safety failure cascades from perception

In October 2023, a Cruise robotaxi dragged a pedestrian twenty feet. It was not a planner failure. It was a perception failure, and every downstream safety guarantee was built on top of it.

safety stack · cascade trace
origin · 01
01perception
scene misread
02planner
acted on what it was shown
03safety oversight
flagged no anomaly
04outcome
harm
trace · san francisco · 2023-10-02verdict · perception
03
research

Perception safety

Research toward a perception layer that downstream safety systems can rely on. Every safety guarantee in an autonomous machine only holds if the perception layer underneath is honest.

04
collaborate

If the thesis matches what you work on, write to us

Open to co-authored research, practitioner interviews, and conversation with safety engineers in humanoid and AV programs. Help us make autonomy safe for the people who will share roads, factories, and homes with these machines.

hunter@protocolz.org