Excited to share that we've been selected for the second cohort of the MassRobotics Physical AI Fellowship! We're proud to join an incredible group of companies pushing the boundaries of Physical AI. If you're heading to GTC, let's connect! #PhysicalAI #Robotics #MassRobotics

7/ Check out how models trained through our pipeline perform across additional tasks on our YouTube channel: 🔗 https://t.co/mF71EKDwYR 🤖🦾🤖🦾

6/ See the resulting policy after two rounds of online-data driven improvement on the popcorn-serving task

5/ [2] Second, adapt the foundation model to a target robot with focused teleoperation data—then improve it in a closed loop. We improve models through online rollouts, iteratively refining both the data strategy and policy performance. The process typically takes approximately

5/ The resulting large-scale dataset, developed under this philosophy, is used to train CFG-1, our first-generation pretrained foundation model for bimanual manipulation:

4/ To ensure data diversity, we systematically broaden variations in environments, objects, and actions during data collection for each target task. The video below showcases our internal explorer and the diverse scenarios generated to date:

3/ As we scale throughput of the data, we prioritize its quality—defined by both precision and accuracy of action, as well as diversity across tasks, environments, and objects. The video below qualitatively demonstrates the accuracy and precision of the estimated 7-DoF,

2/ Our system is built on two core ideas: [1] First, start with humans to scale action data quickly and economically Through our in-house data pipeline, we currently collect approximately 20k hrs/month of action data (~100k hrs collected to date). Each scenario in the figure

1/ At a high level, our system is designed to enable rapid adaptation to new tasks and target robots (see our Product page (🔗 https://t.co/A7bH3n1vW7) for an overview). The figure below illustrates a core aspect of this end-to-end workflow:

Hello world 🤖👋🏻—We are Config. Today, we’re excited to share a preview (🔗 https://t.co/ze9pR37efQ) of what we’ve been building. Our mission is to make robots capable of reliably performing two-handed tasks across diverse real-world settings materially more cost- and