Changelog

Viam has added support for the TensorFlow, PyTorch, and ONNX ML model frameworks, expanding upon the existing support for TensorFlow Lite models. You can now upload your own ML model using any of these frameworks for use with the Vision service.

You can now use the ultrasonic sensor component with the Micro-RDK to integrate a HC-S204 ultrasonic distance sensor into a machine running the Micro-RDK.

You can now edit the configuration of an existing machine that has been configured with a fragment by using the fragment_mods object in your configuration. You can use the fragment_mods objects to be able to deploy a fragment to a fleet of machines, but still be able to make additional per-machine edits as needed.

You can now use the dual GPS movement sensor component to integrate a movement sensor that employs two GPS sensors into your machine. The dual GPS movement sensor calculates a compass heading from both GPS sensors, and returns the midpoint position between the two sensors as its position.

You can now use the Viam Agent to provision your machine or fleet of machines during deployment. The Viam agent is a software provisioning manager that you can install on your machine which manages your viam-server installation, including installation and ongoing updates, as well as providing flexible deployment configuration options, such as pre-configured WiFi network credentials.

You can now use the generic service to define new, unique types of services that do not already have an appropriate API defined for them.

You can now upload machine learning (ML) models to the Viam registry, in addition to modules. You may upload models you have trained yourself using the Viam app, or models you have trained outside of the App. When uploading, you have the option to make your model available to the general public for reuse.

Viam has added a sensor-controlled base component model, which supports a robotic base that receives feedback control from a movement sensor.

You can now visualize your data using many popular third-party visualization tools, including Grafana, Tableau, Google’s Looker Studio, and more. You can visualize any tabular data, such as sensor readings, that you have synced to the Viam app from your machine.

See Visualize data with Grafana for a full walkthrough focused on Grafana specifically.

You can now configure webhooks to trigger actions when certain types of data are sent from your machine to the cloud.

Viam has added a filtered-camera module that selectively captures and syncs only the images that match the detections of an ML model. For example, you could train an ML model that is focused on sports cars, and only capture images from the camera feed when a sports car is detected in the frame.

Check out this tutorial for more information.

You can now run viam-server on a Raspberry Pi 5 with the new board model pi5.

Users can now have access to different fleet management capabilities depending on whether they are an owner or an operator of a given organization, location, or machine.

You can now use API keys for authentication. API keys allow you to assign the minimum required permissions for usage. Location secrets, the previous method of authentication, is deprecated and will be removed in a future release.

Once you have added the data management service and synced tabular data, such as sensor readings, to the Viam app, you can now run queries against both captured tabular data as well as its metadata using either SQL or MQL.

For more information, see Query Data with SQL or MQL.

After you upload and train a machine learning model, you can test its results in the Data tab.

This allows you to refine models by iteratively tagging more images for training based on observed performance.

For more information, see Test classification models with existing images in the cloud.

To use this update, the classifier must have been trained or uploaded after September 19, 2023. The current version of this feature exclusively supports classification models.

The Viam platform now supports the Revolution Pi line of PLCs from KUNBUS in the form of a module. This collaboration allows you to leverage the Raspberry Pi-based Revolution Pi, which runs on Linux and has a specially designed I/O modules for streamlined interaction with industrial controls, eliminating the need for additional components.

Read the Viam PLC Support blog post for a step-by-step guide on using a PLC with Viam.

The Modular Registry enables you to use, create, and share custom modules, extending the capabilities of Viam beyond the components and services that are natively supported.

You can:

• Publish modules on the registry
• Add modules to any machine’s configuration with a few clicks
• Select the desired module version for deployment, make changes at your convenience, and deploy the updates to a single machine or an entire fleet.

You can use a mobile application, available for download now in the Apple and Google Play app stores, to connect to and control your Viam-powered machines directly from your mobile device.

You now have the capability to use a power sensor component to monitor the voltage, current, and power consumption within your machine’s system.

Viam has written a module that allows you to filter data based on specific criteria before syncing it to Viam’s cloud. It equips machines to:

• Remove data that is not of interest
• Facilitate high-interval captures while saving data based on your defined metrics
• Prevent the upload of unnecessary data

To learn more, see this tutorial on creating and configuring a data filtration module.

You can now use boards like the Mediatek Genio 500 Pumpkin that run Linux operating systems with the customlinux board model.

This update enables you to get a closer examination of your image and streamline your image annotation experience by making it easier to add bounding boxes and labels in the Data tab.

With the latest improvements, you can now:

• Navigate between images using the arrow keys in the main image view
• Expand images for a more detailed inspection by clicking the expand button on the right image panel
• Move between full-screen images effortlessly with the <> arrow buttons or arrow keys
• Return to the standard view by using the escape key or collapse button

You now have the ability to duplicate any config component, service, module, remote, or process.

To use this feature:

• Click on the duplicate component icon at the top right of any resource
• Optionally, you can modify the component name to distinguish it
• Adjust any attributes, such as motor pin numbers

Viam now supports sign-up/log-in through Apple Single Sign-On.

Note that currently, accounts from different SSO providers are treated separately, with no account merging functionality.

You can now view your sensor data directly in the Viam app to verify data creation and accuracy. If you depend on sensor data to plan and control machine operations, this feature increases access to data and supports a more efficient workflow.

The Python SDK now includes sessions, a safety feature that automatically cancels operations if the client loses connection to your machine.

Session management helps you to ensure safer operation of your machine when dealing with actuating controls. Sessions are enabled by default, with the option to disable sessions.

You can integrate and control ODrive motor controllers with Viam using the odrive module from the Viam registry.

See the Odrive module readme to learn how to connect and use an ODrive motor controller with Viam, and view the sample configurations.

When prototyping a robotic arm, you can now facilitate movement without creating your own motion planning. This update enables you to implement custom models of an arm component as a modular resource by coding three endpoints of the Arm API:

• getJointPositions
• movetoJointPositions
• GetKinematics

Then, use the motion planning service to specify poses, and Viam handles the rest.

For more information, see this tutorial on creating a custom arm.

You can now apply a crop transform to the views of your connected cameras in the Viam app.

This feature enables you to focus on a specific area of your camera feed.

For example, crop a video stream of a busy street to just the sidewalk.

The beta release of the TypeScript SDK allows you to create a web interface to work with your machine, as well as create custom components and services.

You now have the capability to directly train object detection models in addition to image classification models from within the Viam app.

This update allows you to:

• Add labels by drawing bounding boxes around specific objects in your images or a single image.
• Create a curated subset of data for training by filtering images based on labels or tags.

Now when you invite collaborators to join your organization, you can assign permissions to members by setting one of these roles:

• Owner: These members can see and edit every tab on the machine page, as well as manage users in the app. This role is best for those on your team who are actively engineering and building machines.

• Operator: These members can only see and use the remote control tab. This role is best for those on your team who are teleoperating or remotely controlling machines.

For more information about assigning permissions and collaborating with others on Viam, see Fleet Management.

You can now train and deploy image classification models with the data management service and use your machine’s image data directly within Viam. Additionally, you can upload and use existing machine learning models with your machines. For more information on using data synced to the cloud to train machine learning models, read Train a model.

A new parameter, constraint, has been added to the Motion service API, allowing you to define restrictions on the machine’s movement. The constraint system also provides flexibility to specify that obstacles should only impact specific frames of a machine.

You can now access fragments in your machine setup. The configurations you added will now show up automatically in the Components or Services subtabs in the Builder view. This makes it easier to monitor what fragments you’ve added to your machine and how they’re configured.

For more information, see Fragments.

You now have the option to reuse a machine config from a previous Try Viam session.

The Viam app Code sample tab now dynamically updates as you add resources to your machine’s config. The code samples instantiate each resource and include examples of how to call a Get method on it.

Find more information in the TypeScript SDK docs.

When adding frames to your machine’s config in the Viam app, you can now use the Frame System subtab of the CONFIGURE tab to more easily visualize the relative positions of frames.

Micro-RDK is a lightweight version of the RDK that can run on an ESP32. Find more information in the micro-RDK documentation.

On your machine’s CONTROL tab on the Viam app, you can now set the power of a base. The base control UI previously always sent 100% power to the base’s motors.

The AMS AS5048 is now supported.

The movement sensor API now includes a GetLinearAcceleration method.

The motion service now supports capsule geometries.

The UR5 arm model has been improved using this new geometry type.

You can now implement your own custom resources as modular resources.

You can now supply kinematic information using URDF files when implementing your own arm models.

There are two new movement sensor models:

• ADXL345: A 3-axis accelerometer
• MPU-6050: A 6-axis accelerometer and gyroscope

A new servo model called gpio supports servos connected to non-Raspberry Pi boards.

A badge in the Viam app now displays RTT (round trip time) of a request from your client to the machine. Find this indicator of the time to complete one request/response cycle on your machine’s CONTROL tab, in the Operations & Sessions card.

The Python SDK now supports Python 3.8, in addition to 3.9 and 3.10.

A new API method parameter, extra, allows you to extend modular resource functionality by implementing the new field according to whatever logic you choose. extra has been added to the following APIs: arm, data management, gripper, input controller, motion, movement sensor, navigation, pose tracker, sensor, SLAM, vision.

viam-server now initializes and configures resources in the correct order. For example, if the SLAM service depends on a LiDAR, it will always initialize the LiDAR before the SLAM service.