Developer makes Steam Controller drive itself back to its charging puck

Ray Foss, an aerospace worker and programmer, has built a web app that steers Valve’s Steam Controller back to its charging puck after you set it down on a desk.

Foss calls the project Auto-Charge Vision Tracker. A user opens the web app in a browser, connects the controller, places a camera above the desk, then marks three points: the charging puck, the front of the controller, and the back of the controller. Foss then uses computer vision to calculate the controller’s position and orientation, and the browser sends haptic commands that make the gamepad crawl toward the puck.

The trick turns the Steam Controller’s vibration system into a crude locomotion system. Valve built those motors for haptic feedback during play, but Foss uses them as actuators. The controller vibrates against the desk surface, and friction converts that vibration into movement. The result looks less like a normal charging accessory and more like a small robot vacuum hunting for its dock.

The setup needs a flat, clean surface. A cable, coaster, keyboard edge, or raised mouse pad can stop the controller before it reaches the puck. The camera also needs a clear overhead view because the app depends on position tracking rather than onboard navigation.

Foss designed the project as a browser workflow, so users avoid a native installer. That browser-first approach fits the modern gamepad hacking pattern: Web APIs can talk to hardware, cameras can feed a vision loop, and hobbyists can prototype control systems without a packaged desktop app. The WebHID API gives compatible browsers a path to exchange reports with human interface devices, although browser support and device permissions still set the limits.

The control problem looks simple until the desk becomes the test track. The app needs to know where the puck sits, where the controller points, and how much each motor command changes the controller’s heading. A stronger vibration pulse can move the shell faster, but the controller can yaw, drift, or stall depending on surface texture. Glossy desks, wood grain, rubber mats, and dust can all change the path.

Foss also flagged wear as a practical issue. Repeated vibration can grind the contact points against the desk and create flat spots on the controller’s shell. Rubber feet would reduce abrasion and could give the controller more predictable traction. That small hardware change matters because the software can only steer within the limits of friction and mass.

The project builds on an earlier web app that let users drive the controller across a desk with rumble commands. Foss adds a feedback loop. Instead of asking the user to steer, the camera watches the controller and the app adjusts the movement path toward the dock.

That feedback loop gives the demo its robotics feel. A normal rumble toy moves until a user stops pressing a control. Foss’ version measures the controller’s pose, compares it with the dock location, then issues new motor commands. The method resembles low-cost robotics projects that use external cameras for localization rather than onboard sensors.

The Steam Controller’s hardware gives modders a useful base. The second-generation controller includes capacitive controls, haptics, motion sensing, and the Puck accessory for charging and wireless connection. Valve sells the controller as part of a broader Steam hardware push that also includes the Steam Machine and Steam Frame.

Valve helped this kind of experimentation by releasing controller and puck CAD files in May 2026 under a Creative Commons BY-NC-SA 4.0 license. That move gives users a path to design accessories, docks, guards, and replacement contact surfaces. A modder who wants rubber feet, a desk-safe skid plate, or a more forgiving charging guide can start from Valve’s geometry instead of measuring the shell from scratch.

The commercial signal matters. Valve does not need a self-driving controller to sell gamepads. The value comes from a community that treats the hardware as a platform. Each visible hack gives the controller more mindshare and gives accessory makers a hint about demand for docks, protective feet, desk mats, and 3D-printed guides.

The limits remain clear. The controller cannot move from a living room table to a TV stand. It cannot climb over clutter. The charging puck still has to sit on the same flat surface within camera view. Users who expect appliance-grade docking will run into the difference between a robotics demo and a finished consumer feature.

Foss’ project still shows why Valve’s open accessory posture matters. The company can ship a controller with haptic motors and a puck. Users can connect a camera, write a control loop, add a rubber-foot mod, and turn the same parts into a desk robot. That exchange gives Valve a hardware ecosystem with more surface area than the spec sheet suggests.