How do you turn human motion into useful robot data?
@pzrsaa built a glove-controlled teleoperation stack at Yaak, then used it to collect 221 robot demonstrations.
He wrote up how it works: tracking, calibration and inverse kinematics
I built a teleop stack that lets me control a robot arm with mocap gloves
if you're curious about how it works: parsam.io/gloves/
I ended up using it to collect 221 robot demonstrations. the difficult parts were tracking the glove in space, mapping it onto a clanker, and keeping everything stable enough to collect useful data
Building autonomy for one embodiment used to take years, now imagine doing it for dozens.
Pairing Yaak’s embodiment-agnostic hardware and Nutron (our data platform) with copper-rs (the Physical AI OS) lets us build intelligent machines as fast as software.
One stack. Any embodiment.
How Copper is accelerating our path to safe and scalable autonomy: yaak.ai/blog/yaak-x-copper
Yaak AI retweeted
I've taken over the @Yaak_AI account, what do people want to see related to robotics?
Intelligence for the physical world has to come from the physical world.
We built a Spatial Intelligence Kit that pairs with any forklift, pallet jack, loader or sweeper in a few hours.
Full thesis → yaak.ai/blog/pair-learn-auto…
Intelligence for the physical world has to come from the physical world.
We built a Spatial Intelligence Kit that pairs with any forklift, pallet jack, loader or sweeper in a few hours.
Full thesis → yaak.ai/blog/pair-learn-auto…
Natural language search on #Nutron: “Harsh braking on the motorway”. 7 scenarios found in the open L2D dataset hosted on Hugging Face—the largest dataset for autonomous driving research.
Try it now: eu1.hubs.ly/H0hNQBN0
#AI #OpenSource #Autonomous
Natural language search on #Nutron: “Make a multilane right turn at the traffic lights”. 22 scenarios found in the open L2D dataset hosted on @huggingface the largest dataset for autonomous driving research.
Try it now: eu1.hubs.ly/H0hM2X90
#AI #OpenSource #Autonomous
🚗 We’re excited to partner with @Deloitte to develop a new methodology for safety validation in autonomous vehicles! Together, we’re enhancing virtual testing and safety standards.
eu1.hubs.ly/H0hP0xw0
#Yaak #Deloitte #SelfDriving #AutonomousVehicles #SafetyValidation
Natural language search on #Nutron: “Give way to any oncoming traffic, then enter the roundabout”. 293 scenarios found in the open L2D dataset hosted on @huggingface—the largest dataset for autonomous driving research.
Try it now: eu1.hubs.ly/H0hMxbj0
#AI #OpenSource #ML
Yaak AI retweeted
5000+ hours of fully labeled driving data, released by German startup @Yaak_AI w/ Huggingface.
A banger just got released 💥
Here is a snapshot of L2D, the biggest self-driving dataset by far!
- 90 TeraBytes of data
- 5000 hours of driving
- 6 surrounding HD cameras
- OPENLY AVAILABLE
- Train your car to drive like @Tesla at home
🧵 More details in thread
Yaak AI retweeted
@Yaak_AI’s collab with @huggingface on L2D is so damn cool
A banger just got released 💥
Here is a snapshot of L2D, the biggest self-driving dataset by far!
- 90 TeraBytes of data
- 5000 hours of driving
- 6 surrounding HD cameras
- OPENLY AVAILABLE
- Train your car to drive like @Tesla at home
🧵 More details in thread
Natural language search on #Nutron: "Harsh braking from a speed above 20". 270 scenarios found on the L2D dataset on @huggingface. The biggest open dataset for self-driving!
Try it yourself at eu1.hubs.ly/H0hBJFk0
Yaak AI retweeted
A banger just got released 💥
Here is a snapshot of L2D, the biggest self-driving dataset by far!
- 90 TeraBytes of data
- 5000 hours of driving
- 6 surrounding HD cameras
- OPENLY AVAILABLE
- Train your car to drive like @Tesla at home
🧵 More details in thread
We are at #roscon2024 this week. Thanks @clearpathrobots for stopping by with the Husky explorer.
