Weekly supplier brief
What changed this week, and what to check before your next collection.
The last seven days
2026-09-29 – 2026-10-06 · UTC
Source-linked developments, grouped by collection relevance.
Human video with action supervision
Collection consideration: Plan wrist and hand-action annotations, camera calibration and alignment with the target robot. Ordinary RGB footage alone does not provide the supervision used in this study. Research basis ↗
Robotics — Paper and dataset web discovery · · First discovered; source date unavailable
Physical AI at Scale: Why Robotics Needs a New Data Infrastructure
The paper explores the challenges and opportunities in scaling Physical AI for robotics, focusing on the need for real-world data, the limitations of simulation, and the importance of a continuous data flywheel. It discusses the role of reinforcement learning, human demonstration learning, and Edge AI in enabling scalable robotic deployments. Qualcomm's capabilities in XR, edge computing, and AI infrastructure are highlighted as key enablers for this emerging field.
Original source ↗arXiv Robotics — research abstracts ·
SoTa: Soft Tactile Skins for Dexterous Manipulation
This research presents SoTa, a low-cost capacitive tactile skin that provides full-hand coverage for both humans and robots. The sensor features a shared layout of 202 taxels across finger and palm regions, enabling human-robot co-training with a common tactile encoder. Tactile observations improve success rates in dexterous manipulation tasks, with human demonstrations more than doubling mean success across eight evaluation conditions.
Original source ↗Skild AI — Blog · · First discovered; source date unavailable
Learning from Human Videos for Robotics
Skild AI proposes using human video data to overcome robotics' data bottleneck. By observing human actions, robots can learn new tasks with minimal direct interaction.
Original source ↗arXiv Robotics — research abstracts ·
Towards a General Humanoid Loco-Manipulation Model via Egocentric Whole-Body Human Data Pretraining
The paper presents HumanVerse-500, a 500-hour dataset of human loco-manipulation behaviors collected with a lightweight wearable system. It introduces λ₀, a whole-body humanoid vision-language-action policy trained through three stages, achieving state-of-the-art performance on real-world tasks and analyzing how human data supports downstream control.
Original source ↗
Question to resolve: Which action representation can you deliver, and who will provide the paired robot data?
How this brief works
This rolling brief uses the latest published research, model, dataset and simulation coverage, up to 80 reports. Topic matches are reading aids. They do not prove a model used a dataset or that a buyer is purchasing data.
Collection considerations are RoboSignal’s interpretation of the cited research. They are not new findings from every linked report.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).