Roman Shtylman

Computer Whisperer

Co-founder @

[Foxglove @foxglove]

🚨 We just gave the Plot panel a major tune-up. This wasn’t a redesign, but a focused update to remove friction and improve clarity across the board.

Highlights include:
• New zoom and axis navigation tools
• Point inspection and measurement across plot types
• 1:1 axis locking for spatial XY plots
• Cleaner config with clearer time and axis options
• Improved tooltips and tick alignment
• Lower memory usage for high-density plots

If plots are core to your workflow, these changes are for you.

check out the full blog 👉 https://t.co/xfIanrHSBu

[Roman Shtylman @defunctzombie]

@jorilallo That's the new staff software engineer.

[Roman Shtylman @defunctzombie]

@adrianmacneil One of these is not like the other.

[Roman Shtylman @defunctzombie]

@adrianmacneil The answer is none.

[Foxglove @foxglove]

The MIN3D dataset visualized using Foxglove.

Developed by the 3DOM-FBK research group, the MIN3D dataset captures subterranean scenarios using a custom-built wheeled robotic platform. These environments are characterized by unstructured terrain, complex geometry, poor visibility, and constrained navigation.

To tackle these challenges, the robot was outfitted with a rich suite of synchronized and calibrated sensors, including Basler stereo grayscale cameras, Intel RealSense RGB and depth sensors, a FLIR thermal infrared camera, Livox Horizon and actuated Velodyne VLP-16 LiDARs, and multiple IMUs (RealSense, Livox, and NGIMU). This comprehensive sensor fusion enables data-driven development across visual-inertial SLAM, lidar-based mapping, thermal-aware perception, and multimodal learning.

Visualized in Foxglove, the dataset showcases tightly synchronized data streams—stereo imagery, RGB-D, dense and wide-FOV LiDAR point clouds, thermal imagery, and IMU data—allowing researchers to explore the full depth of sensor data in one unified platform.

MIN3D is particularly well-suited for researchers and developers building resilient autonomy solutions in GPS-denied and sensor-degraded environments, offering a valuable benchmark for SLAM, mapping, and navigation systems under extreme conditions.

This dataset is thanks to the contributions of Trybała, Paweł and Szrek, Jarosław and Remondino, Fabio and Kujawa, Paulina and Wodecki, Jacek and Blachowski, Jan and Zimroz, Radosław. Link to the dataset 👉 https://t.co/P6IsdN648l

[Roman Shtylman @defunctzombie]

@adrianmacneil My first step of any robotics hardware purchase is to wipe any existing OS/install and replace it with my own builds.

[Roman Shtylman @defunctzombie]

@jorilallo Every day this feature makes me hate MacOS.

[Roman Shtylman @defunctzombie]

@adrianmacneil This chart is incorrect. Both `o3` and `o4` have the same amount of `o`.

[Foxglove @foxglove]

🚨 UX update: expand or trim the visualized time-range ↔️

With release 2.20, you can now refine the range of data visualized in Foxglove with ease. When replaying a recording or a specific time range of data, simply click the “Adjust playback range” button in the lower left corner of the screen to make updates.

This feature allows you to:
• Quickly adjust the range by dragging the handles on the playback bar.
• Enter specific times into the dialog for precise control over your selection.
• Copy a link to share your updated time range with your team, enabling seamless collaboration.

[Roman Shtylman @defunctzombie]

@adrianmacneil wut

[Foxglove @foxglove]

📣 Announcing: the Foxglove Academic Plan 🧑‍🎓

Behind every robotics and embodied AI breakthrough are the researchers, educators, and students reshaping how machines interact with the world. 💜

To support these innovators, Foxglove is launching a free Academic Plan for academic and nonprofit teams. Built to address tight budgets, limited access to professional tools, and collaboration challenges, the plan delivers cutting-edge visualization, data management, and team tools—empowering you to streamline workflows and accelerate progress in robotics and embodied AI.

Cheers to the entire academia community and a huge thank you for all your efforts and breakthroughs!

Link to the full blog and to sign up 👉
blog: https://t.co/xBIk1IYu3q
sign up: https://t.co/OSHlGGkD3Q

[Foxglove @foxglove]

🚨 Foxglove release 2.20 is now available 🚨

You can now expand or trim the data range visualized in Foxglove by clicking the “Adjust playback range” button in the lower-left corner while replaying a recording or time range of data.

The topics sidebar has been enhanced with a search feature, making it easier to narrow down to specific topics and expand them to view the fields within their message schema.

The Log panel has been completely rewritten for improved performance. It now includes indicators in the scrollbar, helping you quickly locate warning and error messages.

Additionally, you can now copy links to recordings directly from the recording details page. This makes sharing important recordings with your colleagues seamless, whether you’re using a browser or the Foxglove desktop app.

This release is packed with even more new features, improvements, and fixes. Check out the changelog and download Foxglove v2.20 today.

Changelog: https://t.co/3aDkhNybHa
Download: https://t.co/Qd7HoAuPON

[Foxglove @foxglove]

🧑‍💻 Converting the @Waymo Open dataset to MCAP 👇

Datasets play a pivotal role in advancing robotics and embodied AI, providing essential material for developing, evaluating, and refining algorithms used in perception, decision-making, and control. Yet, the diversity in dataset formats poses significant challenges, as many are designed with unique structures suited to specialized applications, making seamless integration difficult.

#MCAP offers a universal, efficient format for storing and sharing multimodal data. Its standardized structure simplifies data organization, streamlines workflows, and ensures compatibility across tools, making it easier to manage. MCAP’s seamless integration with Foxglove further enhances visualization and analysis capabilities.

In this post, we’ll demonstrate how to convert the Waymo Open dataset to the MCAP format and visualize it using @Foxglove. Check out the full tutorial, including the complete example code 👉 https://t.co/osAoec42Zf

[Foxglove @foxglove]

The @1x_tech #EVE humanoid robot visualized using @Foxglove 🤖

EVE, a humanoid robot standing 6 feet 2 inches tall and weighing 192 pounds, is engineered for industrial and domestic environments with human-inspired musculature driven by Revo1 direct-drive motors. Equipped with multi-terrain wheels and a carrying capacity of 33 pounds, EVE operates for up to six hours on a single charge, with a recharge time of just one hour. Powered by an Intel i7 processor and Nvidia Xavier module running on a custom Linux-based OS, EVE combines real-time control with advanced AI capabilities.

Central to EVE’s intelligence is the 1X World Model, a learned simulator trained on thousands of hours of real-world data. This model enables predictive simulations for complex tasks involving rigid, deformable, and articulated objects, such as opening doors or manipulating items. EVE utilizes the model for long-horizon task planning, simulating various action outcomes to adapt dynamically to its environment. This enhances precision, safety, and efficiency while enabling autonomous navigation and interaction in complex physical spaces.

The 1X World Model also supports embodied learning, allowing EVE to learn new tasks by observing humans and practicing in simulations. This scalable approach ensures continuous improvement in task performance. With potential applications in logistics, security, and beyond, EVE is a versatile solution poised for seamless integration into diverse real-world settings.

[Foxglove @foxglove]

The Foxglove NVIDIA Isaac Sim extension enables real-time visualization of robotics simulation data directly in Foxglove.

In case you missed our previous post on the extension, we released an Isaac Sim extension that enables seamless visualization of simulation data in Foxglove. The extension automatically detects all cameras, IMUs, and articulations in the simulation stage, making their data—along with the complete Transform Tree—available in Foxglove.

In this blog, we'll take a deeper look into the extension's code to understand how it works and explore ways it can be expanded upon.

Check it out 👉 https://t.co/a1SBMzKfhU

[Foxglove @foxglove]

🔦 Spotlight: Monash Motorsport autonomous race car and using Foxglove to develop autonomous concepts. 🏎️ 🏁

Following two successful Driverless demonstrations at the Australasia competition in 2022 and 2023, Monash Motorsport set their sights on a new challenge: their first European campaign in six years. This time, they aimed to compete not only in the EV category but also to showcase their fully integrated autonomous system on a global stage, participating in the two largest Driverless competitions worldwide.

In 2023, Monash Motorsport undertook a transformative shift from operating two separate vehicles—one driven and one autonomous—to a single, integrated race car. This milestone included designing an advanced electrical loom capable of powering both driven and autonomous functionalities, alongside a vehicle controller that seamlessly manages torque and speed for both systems.

Link to the full article 👉 https://t.co/NfiqfrA2sW

[Foxglove @foxglove]

Learn how to easily access your robotics data collected in the field from the comfort of your desk.

📆 Join us on December 10th for a virtual live demo of the Foxglove Agent.

The Foxglove agent simplifies data extraction and management by running directly on your robots, automatically monitoring designated directories for new recordings and securely uploading them to the cloud.

Register for the invite 👇
https://t.co/waMdYwF4BW

[Foxglove @foxglove]

🏎️💨 The AssettoCorsaGym dataset visualized using @foxglove 🏁✨
A simulation benchmark for autonomous racing with large-scale human data.

Built on the high-fidelity racing platform of Assetto Corsa, this dataset provides an unparalleled foundation for developing and testing cutting-edge algorithms in realistic and dynamic environments.

With 64 million data steps, the dataset blends human ingenuity with advanced machine learning. It includes over 2.3 million steps recorded from human drivers of various skill levels—ranging from professional e-sports drivers to casual enthusiasts—alongside data generated by Soft Actor-Critic policies. This unique mix allows researchers to explore human-like decision-making and optimize AI strategies.

Racers can experience four diverse tracks, each offering its own challenges. From the straightforward oval of Indianapolis to the intricate turns of Barcelona (shown here), the balanced mix of Austria, and the high-speed complexity of Monza, these tracks test algorithms under a variety of conditions. Three distinct cars—the nimble Mazda Miata NA, the sleek Dallara F317, and the powerful BMW Z4 GT3—add further depth, ensuring robust testing across different performance profiles.

The dataset is more than just numbers—it’s a tool for researchers to use to benchmark their autonomous driving algorithms, whether they focus on reinforcement learning, model predictive control, or other techniques. By replicating real-world scenarios, AssettoCorsaGym bridges the gap between simulation and reality, making it an invaluable resource for advancing the field.

This dataset is thanks to the contributions of Adrian Remonda, Nicklas Hansen, Ayoub Raji, Nicola Musiu, Marko Bertogna, Eduardo E. Veas, and Xiaolong Wang. 🙏

Link to the dataset 👉 https://t.co/FMPWsQ3Rl0

[Foxglove @foxglove]

🚘 The comma2k19 dataset visualized using Foxglove 🚘

Comma2k19 offers a rich collection of real-world driving data, captured over 33 hours on California’s Highway 280. This dataset is crafted to fuel innovation in developing fused pose estimators and mapping algorithms.

The dataset includes synchronized data from road-facing cameras, GPS, thermometers, and 9-axis IMUs, giving an in-depth view of real-world driving conditions. It also provides raw GNSS measurements for creating precise localization algorithms, along with CAN data that reveals vehicle dynamics essential for advanced control systems.

Technically, the dataset is organized into 10 chunks, each containing about 200 minutes of driving data, with processed logs in numpy arrays and global camera poses. It supports sensor fusion, enabling tightly coupled INS/GNSS/Vision optimizers for robust pose estimation. Data is available in raw and processed formats, with standardized timestamps and sensor units.

Given the above, we added pose projection directly onto images in a dedicated image annotation topic, to visualize the car’s movement and spatial orientation within the frame. This addition is invaluable for building more accurate visual odometry, pose estimation, and scene reconstruction algorithms, as it links 3D positional data with 2D visual contexts, enhancing algorithmic reliability in autonomous navigation systems.

Thanks for the awesome dataset, @comma_ai.

Link check out the dataset 👉 https://t.co/V4sL9X8j7J