Eka Robotics Wants to Be the ChatGPT Moment for Robots
Eka Robotics put a robotic claw on a stage that screws in light bulbs and handles raw food without crushing it. Wired calls it the ChatGPT moment for robotics. The framing is overcooked, the underlying bet is not.
The company was founded in 2025 by Pulkit Agrawal — MIT EECS associate professor, IROS Toshio Fukuda Award winner — and Tuomas Haarnoja, the ex-Google DeepMind robotics researcher who shipped most of the high-profile soccer-robot work. Cambridge, MA, $13M raised from E14 Fund and friends. The model is what they call Vision-Force-Action — VFA, not the more common VLA. The difference matters: they're putting force feedback into the same loop that consumes vision and emits motor commands. The grippers are custom and have a sense of touch. The pitch is that touch is the channel that breaks the speed-versus-generality tradeoff that's been holding manipulation back.
The interesting part is the bet shape. Figure and 1X are pouring capital into humanoids. Anduril and Helsing are weaponizing autonomy. Eka is going narrow on the gripper plus a foundation model that's specifically trained for dexterous, force-aware manipulation. If the bet is right, you skip the humanoid form-factor question entirely — you sell capable hands into factories, kitchens, and stockrooms long before anyone agrees on whether the robot should walk on two legs.
This is the second academic-led VFA-style bet to put real capital on the table in a month. The robotics agent stack is splitting into hardware-first (Figure, 1X), simulation-first (NVIDIA Isaac), and sense-action-first (Eka, Physical Intelligence). The third bucket is the one most agent people aren't watching. They probably should.
https://www.wired.com/story/ekas-robotic-claw-feels-like-were-approaching-a-chatgpt-moment/
← Back to all articles
The company was founded in 2025 by Pulkit Agrawal — MIT EECS associate professor, IROS Toshio Fukuda Award winner — and Tuomas Haarnoja, the ex-Google DeepMind robotics researcher who shipped most of the high-profile soccer-robot work. Cambridge, MA, $13M raised from E14 Fund and friends. The model is what they call Vision-Force-Action — VFA, not the more common VLA. The difference matters: they're putting force feedback into the same loop that consumes vision and emits motor commands. The grippers are custom and have a sense of touch. The pitch is that touch is the channel that breaks the speed-versus-generality tradeoff that's been holding manipulation back.
The interesting part is the bet shape. Figure and 1X are pouring capital into humanoids. Anduril and Helsing are weaponizing autonomy. Eka is going narrow on the gripper plus a foundation model that's specifically trained for dexterous, force-aware manipulation. If the bet is right, you skip the humanoid form-factor question entirely — you sell capable hands into factories, kitchens, and stockrooms long before anyone agrees on whether the robot should walk on two legs.
This is the second academic-led VFA-style bet to put real capital on the table in a month. The robotics agent stack is splitting into hardware-first (Figure, 1X), simulation-first (NVIDIA Isaac), and sense-action-first (Eka, Physical Intelligence). The third bucket is the one most agent people aren't watching. They probably should.
https://www.wired.com/story/ekas-robotic-claw-feels-like-were-approaching-a-chatgpt-moment/
Comments