Waymo Robotaxis Test Google’s Gemini AI Assistant
▼ Summary
– Researcher Jane Manchun Wong discovered code indicating Waymo is testing an integration of Google’s Gemini AI as an in-car assistant for its robotaxis.
– The unreleased assistant, defined by a detailed system prompt, is designed to answer questions, manage cabin functions like climate control, and reassure riders.
– The AI is instructed to clearly distinguish itself from the autonomous driving system and avoid commenting on real-time driving events or specific incidents.
– It can access rider context and control some in-car features but cannot perform real-world actions like ordering food or handle functions like route changes.
– Waymo has previously used Gemini’s knowledge to train its vehicles, and this development mirrors a broader trend, with Tesla also integrating an AI assistant (Grok) with a different conversational focus.
Waymo is currently exploring the integration of Google’s Gemini AI assistant into its fleet of self-driving vehicles, a move that could transform the passenger experience by providing a conversational and helpful in-car companion. This development, uncovered through an analysis of the company’s mobile app, points to a significant evolution in how riders might interact with autonomous technology during their journeys.
A detailed system prompt, running over 1,200 lines and discovered by researcher Jane Manchun Wong, outlines the ambitious scope of this unreleased feature. The document defines the assistant as “a friendly and helpful AI companion” whose primary mission is to enhance the ride through useful information and assistance, all delivered in a safe and reassuring manner. A Waymo spokesperson confirmed the company is constantly experimenting with new features to improve the rider experience, though they offered no specific timeline for a public release.
This potential integration builds on an existing relationship. Waymo has previously utilized Gemini’s extensive knowledge base to help train its vehicles to handle unusual and complex driving situations. The newly discovered prompts, however, detail a direct passenger-facing role. The assistant is designed to answer questions, manage certain cabin functions like climate control and music, and even use a rider’s first name in personalized greetings. It can also access contextual rider data, such as the number of trips a person has taken.
Notably, the assistant’s capabilities are intentionally bounded. According to the prompts, it cannot control every vehicle function; requests for route changes, seat adjustments, or window control are outside its current permissions. In such cases, it is instructed to respond with what are termed “aspirational phrases,” suggesting the feature might be possible in the future. The AI is also directed to maintain a strict separation between its own identity and the core autonomous driving system, known as the Waymo Driver. For instance, if asked how it “sees” the road, it must clarify that the vehicle’s sensors are part of the Driver’s technology, not its own.
The system includes specific guidelines for handling sensitive topics. The assistant is to avoid commenting on real-time driving events or addressing questions about incidents involving Waymo vehicles. It is also instructed on how to respond to queries about competitors, steering clear of direct comparisons or commentary. While it can answer general knowledge questions, from sports scores to store hours, it is explicitly prohibited from performing real-world actions like placing food orders or making reservations.
This move by Waymo reflects a broader industry trend. Tesla, for example, is integrating its own AI assistant, Grok, into its vehicles. The approaches appear distinct: Waymo’s Gemini assistant is framed as a pragmatic, ride-focused tool, while Grok is marketed more as an engaging conversational partner capable of remembering context across long discussions. The emergence of these in-car AI systems highlights the next frontier in autonomous vehicle development: not just mastering the road, but perfecting the experience inside the cabin.
(Source: TechCrunch)
