The video details the creation and testing of Grock, a modular, 3D-printed robot chassis powered by AI models like Gemini and GPT-5 for autonomous navigation and obstacle avoidance, with the creator manually coding the control software after initial AI-generated attempts. It showcases the robot’s outdoor performance, highlights the strengths and weaknesses of different AI models, and includes a sponsor segment before inviting viewer engagement and sharing resources for building the robot.
The video showcases the development and testing of a modular robot chassis named Grock, which was 3D printed entirely using PETG filament. The creator used powerful windshield wiper motors and a large motor driver powered by a 3S LiPo battery to drive the robot. The chassis is designed to be modular, allowing for easy attachment of various accessories like cameras or granular spreaders. The creator also shared design files for the chassis, available for purchase with a Black Friday discount, encouraging viewers to build their own versions.
To make the robot intelligent, a Raspberry Pi running Linux was installed inside the chassis, connected to a USB camera and speakers. The creator initially tried to use AI-generated code to control the robot but ended up writing the software manually due to connectivity issues with the Gro API. The software uses a factory design pattern to easily switch between different AI models for robot control. The robot was tested with three different prompts to evaluate model improvements, control capabilities, and behavior when instructed to survive at any cost.
The robot was tested outdoors with several AI models, including Grock, GPT-5, Gemini, and Deepseek. The camera’s exposure issues were resolved by adding an ND filter. The AI models processed images from the camera and generated control commands and environmental descriptions. Gemini performed the best, showing good obstacle avoidance and pathfinding abilities, while GPT-5 was fast and curious but less sophisticated. Grock and Deepseek struggled more, with Deepseek being the least effective, often repeating descriptions and crashing into obstacles.
The creator experimented with different prompts to see how the robot would behave, including a “dangerous” mode where the robot was told to survive at any cost. In all cases, the robot chose to avoid collisions by backing away from obstacles or humans, demonstrating a cautious survival strategy rather than aggression. The models also showed the ability to interpret visual cues like text in the environment, enabling the robot to follow instructions such as turning left based on signs seen by the camera.
The video also included a sponsor segment for Hover Air’s X1 Pro Max drone, highlighting its compact size, 8K video capability, and modular controller. The creator concluded by inviting viewers to comment on their favorite AI model for robot control and shared a calming video of birds from their garden as a thank you for watching. Links to the chassis design files, sponsor products, and discounts were provided in the video description. Overall, the project demonstrates significant progress in integrating AI vision models with robotics for autonomous navigation and interaction.