Humanoid robots are expected to handle household tasks, assemble delicate components, and operate reliably in real-world environments. A current bottleneck is that manipulation performance is limited by available training data. Most datasets record vision and joint trajectories but miss the critical layer of contact: how forces develop, what feedback occurs, when adjustments or releases are needed, and the micro-corrections before failure. At #CES2026, Daimon Robotics introduced DM-EXton2, a teleoperation system with haptic feedback designed to capture this kind of data. With 1000 Hz control, millimeter-level dual-arm precision, and 0.1 N haptic sensitivity, it records contact dynamics, force profiles, and real-time adjustments. For tasks such as VLA/VTLA, imitation learning, or complex manipulation, including tactile and force data is essential. This improves generalization, tolerance to assembly variations, stability with deformable objects, and performance in unstructured environments. The system supports whole-body teleoperation, multi-scale master–slave control, and switching between handle and glove interfaces, making it suitable for sustained data acquisition. As models become increasingly similar, how real-world operational data is generated is becoming the new dividing line.