What's the Deal with Robotaxi Remote Assistants?
ICYMI: Junko Yoshida and I discuss this controversial topic, and propose a new way of looking at things
As part of my new collaboration with Junko Yoshida, we’ve worked through how so-called “remote assistants” are being used by robotaxi companies.
To be clear, we do not expect robotaxis to have overseas drivers with computer-connected steering wheels who are just remotely teleoperating/driving robotaxis down US city streets. But saying they are not drivers is not quite true either. Remote assistants perform parts of driving tasks such as responding to a robotaxi request for help in deciding whether it is OK to go past a stopped school bus, or whether a traffic light is red. This means they very much have a role to play in safety.
We have a detailed exploration of the issues, a podcast video, and a discussion of the takeaways:
A piece by Junko with perspective and takeaways
A video podcast in which Junko & I discuss this topic, as well as our new collaboration. (This is also the main YouTube link below in this posting.)
Here is an extended video clip from that episode in which I just discuss the role taxonomy but without the other discussions (12 minutes)
A piece by me with a detailed taxonomy of operator/assistant roles: operator (driver), supervisor, decider, responder, standby, and maintainer. Each role has specific roles to play in safety.
Here are some relevant pointers to materials mentioned in the above pieces:
The AVSC guidance on remote operators. I strongly disagree with any notion that remote assistants have no role to play in safety that seems to be based on this guidance. In general, I think this document needs a lot of work to address practical safety concerns. Perhaps SAE ORAD would find my taxonomy a more suitable starting point for a potential future SAE standard on this topic.
Materials from Senator Markey’s investigation. The letters from AV companies protesting that their computer is “responsible” or otherwise is in charge of safety in a way that implies the remote operator has no safety responsibility are really problematic here.
The NTSB preliminary investigation report describing that a remote operator mistake contributed to an incident of a robotaxi going past an active school bus stop when it should not have.
Bonus paper: Missy Cummings: What self-driving companies should learn from drone remote operations, March 2026
Bonus substack posting by me: All Robotaxis Have Remote Drivers
Bonus substack posting by me: Remote Assistants for Autonomous Vehicles
Bonus substack posting by me: Laundering Accountability with Embodied AI (Part 1 | Part 2)
Phil & Junko video podcast below:
Going forward, if you want to see my thought pieces for the collaboration with Junko as soon as they come out, you should subscribe to Phil & Junko on AV Safety.
I’ll continue this personal substack as well. Some weeks I’ll have a summary of the collaboration materials as we end a topic cycle. Other weeks I’ll have an original solo post here. I hope you continue to find my content interesting and useful!
Thank you for putting together all the relevant links into one place. "Remote operators in the Philippines" can easily steal a headline and we can all forget about it. But look deeper, there's a subtle shifting going on as to who is accountable for AV safety.