Embodied-learning research advances on teleoperation, tactile sensing, and multimodal control
A cluster of 2026 robotics research — VR-based whole-body teleoperation, reinforcement-learned lower-body controllers, dexterous-hand retargeting, distributed tactile sensing, and the Humanoid Transformer with Touch Dreaming — marks a coordinated push toward humanoids that learn contact-rich physical skills from richer, multimodal signals.
The common thread is closing the sensing-to-action loop for the whole body. Teleoperation supplies human demonstrations, tactile sensing supplies contact feedback, retargeting maps human hand motion to robot hands, and a multimodal transformer ties it together — a research stack aimed at the hardest part of humanoid control, which is manipulating the physical world reliably rather than walking or posing.
Teleoperation is the data engine underneath. Whole-body VR teleoperation lets humans generate the demonstrations that contact-rich policies need, addressing the fundamental bottleneck in robot learning: real physical data is slow and expensive to collect. Pairing it with world-model 'dreaming' to multiply each demonstration is how the data economics start to work.
The direction is the same convergence visible across embodied AI: fold perception, prediction, and control into one learned system trained on multimodal signals. These papers are the research substrate beneath the year's humanoid deployment milestones — the science that has to work for the robots on warehouse floors to keep getting more capable.
IEEE Spectrum — Video Friday: humanoid robot production, Mars rovers, and more → · Humanoid Press — Humanoid daily: robotics and AI news →