Foxglove (@foxglove) header
Foxglove (@foxglove) profile picture

Visualize and manage multimodal data in one purpose-built robotics development platform.

Joined February 2021
Photo by Foxglove (@foxglove)Photo by Foxglove (@foxglove)Photo by Foxglove (@foxglove)Photo by Foxglove (@foxglove)Photo by Foxglove (@foxglove)Photo by Foxglove (@foxglove)Photo by Foxglove (@foxglove)Photo by Foxglove (@foxglove)Photo by Foxglove (@foxglove)Photo by Foxglove (@foxglove)
1/ New on Foxglove: Automations and Datasets, plus new ways to integrate Foxglove with the tools your team already uses. Find the failure, file the issue, version the training data, and train the next model on exactly that version. All in one platform.
2
2
5
521
Ticket #1847, the dream: “RTK dropped from fixed to float under the awning, the position estimate shifted 20 to 30 cm, and the planner trusted it. MCAP recording attached.” The reality: “Sometime yesterday the robot fell over.” @adrianmacneil on closing that gap at FUSE SF 🧵
1
1
4
322
Physical AI runs on a learning loop: collect, triage and debug, search and curate, train, evaluate, operate. It starts with the data. MCAP is self-describing, write-optimized, and indexed for random access, even when compressed.
1
84
Then the demo: a delivery robot falls on the stairs. Adrian traced it to a foot-placement failure, found similar incidents across recordings, and turned them into a training dataset. Thanks to @pointonenav and @svrobo for hosting. #PhysicalAI #Robotics
66
Foxglove retweeted
My XGO dog can locate a ball, come to it, pick it up and bring to a requested toy (surviving the hard hits with RL policy, and finding the lost route, w/o lidar and nav2) after a single voice command. This physical AI moment feels so cool. Another wow moment - how customisable Foxglove is. 🦾
1
1
5
259
Cool, just did @LeRobotHF SO-101 calibration in Foxglove. Yes, the CLI utility is perfectly fine, but it's nice to see my actual assumed pose pre/post calibration and run it all from one place.
2
2
1
6
329
You are likely using LLMs a lot at work (hopefully not to comment under social posts, though). Here is how you can use Foxglove Automations to automatically create datasets. See the full video: hubs.li/Q04FNqZ40
2
194
100 Franka arm recordings, two Foxglove Automations, 17 minutes from upload to the last Linear ticket. One marks grasp misses. The other checks each miss on camera and files a ticket. 17 of 38 flagged misses were the gripper holding tissue or cloth. hubs.li/Q04FF6320
5
372
Replying to @foxglove
@foxglove trained a bimanual policy on a Trossen Mobile AI sorting citrus. The first version missed because of table height. A good read on why the right data matters. buff.ly/X7JC7tx #PhysicalAI #Robotics #ImitationLearning #RobotLearning #RobotArm
1
3
182
Your code is in git. Your checkpoint has a hash. Your training data is a manifest someone will overwrite next week. Foxglove Datasets keep the list: episodes cut from your recordings, committed as versions that never change.
1
1
286
4/ And new ways to connect Foxglove to your stack: the agent can now reach your issue tracker and code host over MCP, so an automation can file the issue for you. Plus OAuth apps, so you can build on Foxglove with each person’s own access.
1
63
Foxglove retweeted
Imagine debugging your robot from another city. Using @Foxglove, attendees in San Francisco viewed live camera feeds and controlled Vega 1 Pro in Fremont, across separate networks. No separate VPN required. See how we connected it and read the full blog: bit.ly/4zsq7n1 #PhysicalAI #Robotics #Teleoperation #Foxglove #Dexmate #Vega #Humanoid
1
2
1
5
639
Another post for @foxglove: I built a record and replay feature for the ROS 2 powered SO-101, and tested some new Foxglove features. Also, first published on Foxglove's blog. kamathrobotics.com/record-an…
2
2
192
Your robot failed in the field. Now what? At ROSCon Global 2026, our CEO Adrian Macneil showed a practical ROS 2 debugging workflow with MCAP: - Fast partial reads - Reproducible playback - Live introspection plus recorded replay Enjoy the talk! hubs.li/Q04DRnJ_0
1
279
Remember when this was considered the state of the art in Robotics? We certainly do! It feels like we are in a different industry now. Video credit: IEEE Spectrum: hubs.li/Q04Bxpwh0
2
5
526