Every funding round, launch and paper, as it's reported.
Anthropic economists scored about 19,000 O*NET job tasks and concluded that robots can technically perform 74% of physical tasks in the US, yet beat human labour on cost for only 0.3% of tasks. At the historical 3% annual decline in robot prices, the study estimates it would take 40 years for that share to reach 10%.
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months.
Generalist AI said on August 19, 2026 that GEN-1.5 can learn a new manipulation task from one demonstration of 3 to 12 seconds placed in its context, averaging 59% success over 10 tasks with no fine-tuning and 83% after 10 gradient steps on about 5 minutes of data. The company says the ability emerged from more than eight months of pretraining, but calls the tasks simple and the skills brittle, and has named no customers.
NeuralActuator, a model from MIT's Computational Design and Fabrication Group with Amazon, won the Outstanding Systems Paper Award at RSS 2026. It learns how cheap servo motors really behave from motor current, voltage and temperature, and estimates external force with a reported mean error of 0.12 N, against 0.66 N to 1.41 N for classical baselines.
OmniRetarget, a motion retargeting method from Amazon's Frontier AI & Robotics team and university collaborators, won both the Best Conference Paper Award and the Best Paper Award on Robot Manipulation and Locomotion at IEEE ICRA 2026 in Vienna. The method preserves contacts between body, objects and terrain, and the authors used it to generate more than 8 hours of trajectories that trained a Unitree G1 to perform parkour and box carrying in the real world without real-robot fine-tuning.
An 82-author team from Toyota Research Institute, Cornell University and MIT reported in Science Robotics on April 15, 2026 that multitask large behavior models, built on diffusion policy, outperform single-task baselines in blind, randomized trials and need far less task-specific data. The preprint version describes about 1,700 hours of robot demonstrations across more than 500 tasks and 1,800 controlled real-world trials.
Generalist AI released GEN-1 on April 2, 2026, an embodied foundation model pretrained on more than 500,000 hours of human physical interaction data captured with wearable devices and no robot data. The company reports 99% average success on tasks where its previous model reached 64%, about one hour of robot data per task and task times roughly three times faster than the prior state of the art; the results are the company's own and have not been independently replicated.
Figure AI on Jan. 27, 2026 released Helix 02, a robot foundation model that runs a humanoid from camera pixels and fingertip touch down to joint commands at 1 kHz, replacing 109,504 lines of hand-written C++ with a 10-million-parameter learned controller. The company showed a Figure 03 unloading and reloading a dishwasher across a full kitchen in a four-minute, 61-action run it says had no resets and no teleoperation.
Generalist AI introduced GEN-0 on November 4, 2025, a family of embodied foundation models that the company says improves predictably with more physical interaction data. It reported a capacity threshold at 7 billion parameters, models scaled past 10 billion, and a pretraining corpus of more than 270,000 hours growing by over 10,000 hours a week. The findings were published as a company blog post, not a peer-reviewed paper.
Boston Dynamics and TRI showed Atlas completing a long workshop task under a single Large Behavior Model that commands walking and manipulation together. The 450M-parameter Diffusion Transformer runs at 30 Hz on camera, body-state and language inputs and learns new behaviours from VR teleoperation data without new code. No success rates or data volumes were disclosed.
Physical Intelligence on 22 April 2025 published π0.5, a vision-language-action model that drove mobile manipulators through 10- to 15-minute cleaning jobs in three homes absent from its training set. The San Francisco start-up says co-training on data from other robots, web tasks and spoken coaching, rather than more home-robot data alone, is what lets the model generalise, and its own ablations show which ingredients matter most.