@codenameposhan: Dream Labs just came out of stealth with Dream-1 (October 19, 2026). Plug in a two-armed robot it has never seen and it can start doing tasks, no training data from that robot. Here's what you need to know. 1. How it works: Dream-1 is a World Action Model. It sees through a head camera and a camera on each wrist, predicts the video and the actions together, and returns where each gripper should go. Your robot's own IK solver turns that into joint moves. 2. Zero-shot on new bodies: Dream Labs tested it on 27 robots it was never trained on in simulation, plus real ones, even the 18-year-old PR2. Scores relative to YAM, the robot it trained on, range from 91.2% (PiPER) down to 36.6% (Bracket Bot). 3. Independent test: Robocurve scored it 59 on RoboDojo-RC Tier 1, ahead of Opus 5.5 and Gemini 4 Argon at 48, at about $0.15 and 1.4 minutes per task attempt. 4. Price: $3 an hour for Dream-1, $2 for Dream-1-flash. Each call returns 1.6 seconds of actions. Fine-tuning as a service adapts it to your robot, in Dream Labs' examples with as little as 5 minutes of data. 5. Backing: a $90M seed led by B Capital with Kindred Ventures, from a team of former NVIDIA GR00T researchers. The catch: it's a research beta, not validated for safety-critical use. Zero-shot works on bimanual robots with parallel grippers, results vary a lot by robot and IK solver, and fine-tuning still beats zero-shot. Follow @codenameposhan for more AI breakdowns. Sources: Dream Labs' research post (Oct 19, 2026) and the Business Wire launch release. Independent briefing by CodenamePoshan, not a Dream Labs post. Dream Labs' logo and artwork belong to Dream Labs; Robocurve data CC BY 4.0. #DreamLabs #Dream1 #Robotics #RobotLearning #AI #EmbodiedAI

Codename Poshan

Open In TikTok:

Region: US

Tuesday 20 October 2026 04:46:58 GMT