PITTSBURGH, Pa. — Viam will debut BoxBot at IROS 2026, showing a research robot that combines image-based visual servoing with a vision-language-action (VLA) model to open taped cardboard boxes. The conference is scheduled for Sept. 27 through Oct. 3 in Pittsburgh.

BoxBot uses image feedback to guide a cutting tip toward the tape seams on sealed boxes. Viam says the seams are typically 1 to 2 millimeters wide, requiring the robot to continuously adjust its position as it cuts.

The second part of the task uses a learned policy. According to Viam, a fine-tuned SmolVLA model handles the deformable, contact-rich task of peeling open and adjusting the box's four flaps.

The system therefore divides the manipulation task between programmed visual control and learned behavior. Visual servoing handles the relatively precise task of following the tape seam, while the learned policy handles the changing geometry and physical contact involved in opening the flaps.

Viam describes BoxBot as an ongoing research project rather than a commercial product. The company says the VLA model was trained from human demonstrations and fine-tuned for the box-opening task.

The project is built around Viam's robotics software platform, with the company providing the system's software and machine-learning components. Viam says the BoxBot work will be shown at its booth during IROS 2026.

The supplied material does not provide an independently measured success rate, cycle time, comparison with other box-opening systems or evidence of commercial deployment. The demonstration therefore provides a look at Viam's approach to combining conventional visual control with learned manipulation, rather than evidence of a production-ready box-opening system.