Touching with torque enables human-level robotic dexterity.

Achieving human-like forceful manipulation remains a major challenge in robotics because of the lack of critical environmental interaction cues such as collisions, balance, and resistance. We present a torque-angle-pressure (TAP) tactile sensor leveraging magnetic flux density gradients to achieve bidirectional, ultrasensitive (~0.1°, ~0.4 newton-millimeter), and high-linearity (R2 = 0.99) sensing over a wide range (±241.6 newton-millimeter) through a single readout channel. The accurate torque sensing ability provides both force and distance information, bringing the environment into the interaction loop. A TAP-equipped robot can perform vision-free stable object placement and complete a balance beam stacking challenge in just 2.4 seconds with a success rate of 81.5%-both measured metrics surpassing human performance. It also supports adaptive daikon slicing with real-time posture and motion adjustments-capabilities rarely achievable in existing robotic systems. This work advances tactile sensing, enables forceful manipulation in unstructured environments, and represents a key step toward effective human-robot collaboration.