What Viam Builds

Viam builds a modular robotics software platform and physical reference designs that connect robot hardware to edge compute and cloud services. The company sells its platform to robotics engineers, commercial operators, and original equipment manufacturers (OEMs). Its software stack provides device management, sensor data capture, real-time telemetry, and machine learning (ML) model orchestration across diverse operating systems. Instead of building isolated custom stacks, operators install the Viam agent onto single-board computers or industrial microcontrollers to control actuators, cameras, and sensors through unified application programming interfaces (APIs). Alongside software, Viam develops research robots like BoxBot to demonstrate end-to-end task automation. BoxBot uses visual servoing and vision-language-action (VLA) models to open packaged boxes without human intervention. The company operates on a software-as-a-service (SaaS) and consumption-based business model. Customers pay for cloud data synchronization, remote data storage, and compute hours used during machine learning workflows. Base client libraries and core open-source tools remain accessible for research teams, allowing low-volume development before commercial fleet deployment.

Viam Robots and Models

These are the Viam models a buyer is most likely to be evaluating, and what each one actually does.

  • BoxBot Research Arm — BoxBot is an automated manipulation research robot designed to open sealed cardboard boxes. It combines an industrial robotic arm with a cutting tool and suction or gripper fixtures to slice tape and fold box flaps. The system relies on image-based visual servoing to locate tape seams measuring 1 to 2 millimeters wide. It uses a vision-language-action (VLA) policy to manipulate cardboard flaps. BoxBot remains an active research platform rather than a turnkey commercial appliance.
  • Viam Robotics Software Platform — The Viam platform is a hardware-agnostic operating architecture that manages robotics components through standardized APIs. It runs locally on edge devices such as the NVIDIA Jetson Orin Nano, coordinating motor drivers, cameras, and environmental sensors. The software handles time-synchronized telemetry capture so operators can record demonstration datasets without writing custom synchronization code. Built-in cloud sync allows engineering teams to push updated machine learning models back to remote machines over wireless connections, reducing the physical maintenance needed for production fleets.
  • SO-101 Teleoperation Workstation — The SO-101 is a leader-follower teleoperation workstation built to collect imitation learning data for robotic arms. The system links a physical input arm directly to a follower manipulation arm, mirroring operator movements in real time. Because the hardware connects directly to the Viam capture layer, joint angles and video feeds sync automatically to cloud storage. This integration enables operators to record hundreds of demonstration episodes without writing custom data logging scripts, accelerating machine learning policy training.
  • Viam ML Model Service — The Viam ML model service is an on-device inference framework designed to run neural network policies at the machine edge. It allows small form-factor computers to execute models like SmolVLA locally, eliminating the latency and reliability risks of cloud round-trips during physical manipulation. Operators configure the service through simple configuration fragments, making it possible to deploy trained vision models or motion policies across multiple robot arms without rewriting native robotic controller drivers.

Viam Pricing and Availability

Viam does not publish a fixed retail price for complete physical machines like BoxBot. Because BoxBot functions as an ongoing research and development project, it is not sold off the shelf as a pre-packaged consumer or warehouse product. Commercial costs depend on the physical robotic arm you specify, the edge computing hardware you select, and the subscription tier you maintain on the Viam software platform. Viam provides a tiered software pricing structure based on connected devices, cloud data storage volume, and computational usage for machine learning pipelines. Core developer modules and open-source packages on Hugging Face are freely accessible for prototyping. For physical implementations, buyers purchase standard hardware components such as an NVIDIA Jetson Orin Nano, industrial robot arms, and compatible cameras from third-party distributors. Integration expenses depend on internal engineering hours or fees paid to systems integrators. Viam operates globally through its cloud services, while hardware lead times match the supply chains of component manufacturers. Enterprise fleets negotiate custom software contracts that include dedicated support, custom service level agreements, and security compliance audits.

Who Viam Robots Are For

Viam suits robotics engineering teams, industrial researchers, and automated logistics developers who need a flexible platform to train, deploy, and manage physical manipulation systems. It serves organizations that possess software engineering capabilities and want to avoid building custom data pipelines for imitation learning and edge inference. BoxBot specifically fits automated warehousing labs and academic institutions studying vision-based box opening, deformable object handling, and end-to-end policy deployment. This platform is not for operators seeking an immediate, plug-and-play appliance that works out of the box without technical configuration. Warehouses that need turnkey, certified packaging decanting systems today should look at established commercial material handling integrators or turnkey providers like Covariant and Dexterity. Organizations without in-house software developers or robotics engineers will struggle to adapt the research codebase to production packaging lines. Buyers with fixed hardware setups that already run deeply integrated legacy programmable logic controllers (PLCs) may find the cloud-centric data model redundant for purely deterministic, repetitive factory floor sorting.

  • Evaluate hardware compatibility before committing. Viam runs across Linux and single-board computers like the NVIDIA Jetson Orin Nano, but legacy factory hardware may require custom driver modules to interface with proprietary industrial programmable logic controllers.
  • Review edge computing requirements for machine learning policies. Running models like SmolVLA locally demands dedicated graphic processing power on the robot, which increases unit component costs, thermal management needs, and electrical power consumption in mobile or compact setups.
  • Check safety certifications and containment needs. BoxBot uses sharp cutting tools to open tape seams, which means deployment on commercial warehouse floors requires physical fencing, light curtains, and compliance with ISO 10218 safety standards for industrial robots.
  • Account for continuous cloud data transmission costs. Streaming multi-camera video feeds and high-frequency joint states to the cloud for policy training consumes substantial network bandwidth, which can raise operational costs in facilities with metered cellular or satellite internet.
  • Assess in-house software engineering capacity. Operating Viam and fine-tuning vision-language-action policies requires familiarity with Python, machine learning workflows on platforms like Hugging Face, and edge deployment tools rather than traditional ladder logic programming.
  • Verify physical box variety and defect rates. Automated box opening relies on image-based visual servoing to track 1 to 2 millimeter tape seams. Damaged cartons, crushed corners, or obscured seams will decrease task success rates without human supervision.

Viam Alternatives

Robots and vendors a buyer typically cross-shops against Viam.

  • Dexterity — Dexterity builds commercial robotics systems for palletizing, depalletizing, and carton handling in high-throughput distribution centers. While Viam provides a software platform and research reference code, Dexterity sells fully supported, turnkey production solutions with complete safety enclosures and continuous operational fleet management services for warehouse logistics operators.
  • Covariant — Covariant supplies universal AI software designed to drive robotic picking, induction, and parcel handling in logistics facilities. Covariant focuses heavily on commercial order fulfillment and warehouse automation, whereas Viam offers a broader, developer-facing robotics development platform that allows engineers to build diverse autonomous systems from scratch.
  • Robot Operating System (ROS) — The open-source Robot Operating System, maintained by the Open Source Robotics Foundation, is the legacy standard for academic and commercial robotics development. ROS provides deep community drivers and simulation tools, but requires teams to build their own cloud synchronization, fleet deployment, and remote data collection tools that Viam includes natively.
  • NVIDIA Isaac — NVIDIA Isaac provides an enterprise robotics platform featuring GPU-accelerated simulation via Isaac Sim and modular perception libraries. NVIDIA focuses on low-level compute optimization, synthetic training data generation, and edge AI execution, whereas Viam emphasizes fleet management, accessible API wrappers, and cloud data collection pipelines across heterogenous hardware.
  • ABB Robotics — ABB Robotics manufactures industrial robotic arms, collaborative robots, and automation software for manufacturing and logistics. ABB delivers hardware backed by global field service, ISO safety certifications, and deterministic controller cabinets, whereas Viam focuses on flexible software, modern machine learning integration, and open developer tooling.

Recent news

A dated log of developments, each tied to its primary source.

  • 2026-09 — Viam introduced BoxBot, a research-stage robot arm that opens taped cardboard boxes using image-based visual servoing to cut tape seams and a fine-tuned vision-language-action model to open the flaps. Trained on 125 teleoperated demonstrations and inferring locally on an NVIDIA Jetson Orin Nano, the project's code and dataset were released openly on Hugging Face. Robotics Tomorrow
  • Release of standardized commercial box-opening hardware kits by third-party integrators using the Viam software stack.
  • Expansion of edge-compatible vision-language-action models optimized for low-power edge processors beyond the Jetson Orin Nano.
  • Publishing of third-party throughput and failure-rate benchmarks for visual servoing on damaged cardboard cartons in live warehouse environments.

Bottom Line

Viam provides a capable software framework for teams building modern robots, demonstrated by research platforms like BoxBot. It reduces the engineering overhead needed to collect demonstration data, train neural policies, and deploy them to edge hardware. While BoxBot remains an open research design rather than a turnkey industrial machine, the underlying architecture offers tangible value to developers creating custom automation. Audit your internal machine learning capabilities and evaluate sample data pipelines before selecting Viam for production fleet deployment.

Explore our comprehensive robot arm buyer guides and warehouse automation profiles to compare hardware platforms for your facility.

FAQs

What is BoxBot?

BoxBot is an automated robotic research platform developed by Viam that opens taped cardboard boxes. The system utilizes image-based visual servoing to locate and cut tape seams measuring 1 to 2 millimeters wide. A fine-tuned vision-language-action policy then opens the cardboard flaps. BoxBot demonstrates how developers can integrate visual perception, imitation learning, and physical motor control on a single hardware controller using the Viam software framework.

Can I buy BoxBot as a commercial product?

BoxBot is currently an ongoing research project and reference architecture rather than a commercial product you can purchase off the shelf. Viam provides the software platform, and open-source models and datasets are hosted on Hugging Face for community development. Organizations wanting to deploy BoxBot must assemble the physical robotic arm, cutting end-effector, cameras, and computing hardware themselves, or collaborate with a systems integration partner to build a custom unit.

What hardware does BoxBot require to operate?

BoxBot operates using an industrial robotic arm equipped with custom cutting and manipulation tools, multi-angle camera sensors, and an edge computing module. In Viam's reference setup, the machine learning models run locally on an NVIDIA Jetson Orin Nano board using the Viam ML model service. The edge device handles real-time camera feeds and motor commands without relying on constant cloud connectivity during the physical cutting and flap-opening sequences.

How is BoxBot trained to manipulate box flaps?

BoxBot is trained through imitation learning using human demonstrations captured with teleoperation controllers. In the initial development phase, two Viam engineers gathered 125 demonstrations over five hours using a virtual reality controller. The Viam platform synced the arm kinematics and dual camera streams to the cloud. The team then fine-tuned an open-source SmolVLA model using LeRobot on Hugging Face Jobs before deploying the policy back to the robot.

How does Viam make money if the code is open source?

Viam generates revenue through a software-as-a-service model focused on cloud data management, telemetry synchronization, and remote machine orchestration. While core developer libraries and reference datasets are free and open source, commercial enterprises pay for cloud storage, continuous fleet monitoring, and remote compute services. As fleets expand from single prototypes to dozens of operational robots, platform usage fees increase according to data ingress, storage duration, and active connected devices.

How does BoxBot handle boxes of different sizes?

BoxBot adapts to varying package dimensions by combining image-based visual servoing with learned neural policies. Visual servoing continuously adjusts the robot arm position based on live camera feedback, guiding the cutting blade along tape seams even if box placement shifts slightly. The vision-language-action policy then processes the visual scene to identify flap boundaries, allowing the robot to manipulate deformable cardboard surfaces without requiring fixed mechanical jigs for every carton size.

Primary Sources