Waymo Trials Gemini In-Car AI Assistant for Autonomous Robotaxis
An autonomous ride-hailing service appears to be testing Gemini as an in-car AI assistant within its robotaxis. According to research findings, the effort aims to add a conversational companion that can accompany riders and respond to their questions.
During an examination of the mobile app code, a full system prompt for an unreleased Gemini integration was discovered. The document outlines in detail how the assistant is expected to behave inside the vehicle. As a result, the findings suggest that the project goes beyond experimentation and reflects a structured internal design.
Although the feature has not appeared in public builds, the system prompt indicates broader ambitions. Rather than acting as a basic chatbot, the assistant can answer questions, manage select in-cabin functions, and reassure riders when necessary. Meanwhile, the company has indicated that it routinely tests new features, some of which may never reach riders.
Design, Capabilities, and Boundaries
Notably, Gemini has already been used internally in other ways. Previously, its general knowledge capabilities helped train autonomous vehicles to navigate complex and high-stakes scenarios. Therefore, the in-car assistant builds on existing AI usage rather than representing a completely new initiative.
The assistant is designed with a defined identity and purpose. It is meant to act as a friendly and helpful AI companion whose goal is to enhance the rider experience in a safe and unobtrusive way. Consequently, it is instructed to use clear language, avoid technical jargon, and keep responses brief.
When activated through the in-car screen, the assistant can deliver personalized greetings and access limited contextual information, such as a rider’s trip history. In addition, it can control certain vehicle features, including temperature, lighting, and music. However, controls for volume, route changes, seat adjustments, and windows remain unavailable. If a rider requests an unsupported feature, the assistant is designed to acknowledge the limitation rather than attempt the action.
Importantly, the system enforces a strict separation between the AI assistant and the autonomous driving technology. As a result, the assistant refers to the driving system in the third person when answering questions about how the vehicle operates.
The prompts also define how sensitive topics should be handled. The assistant must avoid discussing real-time driving actions or specific incidents and should redirect such questions instead of addressing them directly.
Broader Context and Industry Comparison
Despite its limits, the assistant can answer a wide range of general knowledge questions. For example, it may provide information about the weather, landmarks, store hours, or major sports results. However, it cannot perform real-world tasks such as ordering food, making reservations, or responding to emergencies.
At the same time, this development reflects a broader industry trend. Other companies are also embedding AI assistants into driverless vehicles. Even so, the approaches vary. In this case, Gemini appears focused on short, practical interactions tied to the ride experience, rather than extended conversations or long-term memory.
Overall, the testing points to a cautious and controlled approach to in-car AI. As a result, future autonomous rides may include assistants designed to be helpful and informative without overstepping clearly defined boundaries.
