Viam Introduces Innovative Box-Opening Robot at IROS 2026 Conference

Viam Unveils BoxBot at IROS 2026

At the International Conference on Intelligent Robot Systems (IROS) 2026, taking place in Pittsburgh from September 27 to October 3, Viam has introduced a new robotic innovation known as BoxBot. This cutting-edge robot is designed to handle one of the most mundane yet prevalent tasks in numerous industries: opening cardboard boxes.

What is BoxBot?

BoxBot is part of Viam’s ongoing research project aimed at enhancing robotic manipulation capabilities. Unlike traditional methods, this robot employs image-based visual servoing and a vision-language-action (VLA) model, trained on a dataset of human demonstrations, to effectively cut tape and open box flaps. The hybrid approach allows BoxBot to learn and adjust its actions based on image feedback and predefined policies, ensuring precision in a task that varies from box to box.

Key Features and Technologies

Visual Servoing

Viam has equipped BoxBot with sophisticated visual servoing technology, enabling it to identify tape seams that are typically between 1 to 2mm. This capability means that BoxBot can adapt to the subtle differences found in individual boxes, ensuring a clean and efficient opening process.

Learned Policy Approach

Central to BoxBot’s functionality is the SmolVLA model, which has been fine-tuned to perform the complex task of peeling open box flaps. By utilizing learned policies, BoxBot can tackle a contact-rich manipulation task that would otherwise pose challenges for conventional, analytical modeling approaches.

Integrated Operations

What sets Viam’s approach apart is the seamless integration of hardware and software via shared APIs. This allows developers to connect devices using JSON, while the logic can be programmed in Python, Go, or TypeScript. With the ability to run alongside other robotic systems, BoxBot showcases how Viam’s platform can simplify the development process in robotics.

Efficient Data Capture

Viam’s innovative platform has also streamlined the data gathering process required for training BoxBot. During the development phase, Viam engineers utilized a VR controller to capture 125 teleoperated demonstrations in about five hours. This data was synced to the cloud, enabling the team to fine-tune the SmolVLA model with minimal technical burdens, contrasting starkly with older methodologies that often demand extensive engineering efforts.

Open Source Contribution

In a move to foster collaboration in the robotics community, Viam has made BoxBot’s code and training dataset available on Hugging Face. This not only allows other researchers to benchmark their work against BoxBot but also facilitates replication of the setup—promoting innovation within the field.

Live Demonstrations at IROS

Attendees at IROS 2026 will have the opportunity to witness BoxBot in action at Viam’s booth (Booth 828). Visitors can interact with the robot, controlling the arm to gain insights into its operational capabilities. A separate VR teleoperation station will also allow attendees to understand the training process and record new data for further innovation.

Future Directions

Dr. Shannon Bradshaw, SVP of Product and Education at Viam, emphasized the importance of learned policies in tasks that involve intricate physical interactions. As robotics continues to mature, Viam remains committed to pushing the boundaries of what is achievable through its integrated software platform, ensuring that developers can focus on creating functional applications rather than becoming bogged down in technical hurdles.

In conclusion, BoxBot is not just a robotic novelty; it's a glimpse into the future of automation in logistics and beyond. With its sophisticated features and the support of a robust development community, it’s poised to make box-opening a task of the past.

For more information, visit Viam’s website or check out their developments at the IROS 2026 conference.