Grasping is not just a fundamental way in which we interact with objects. It is also a key behaviour that allows us to discover the physical laws governing our environment. An example is how we handle torque, the tendency of objects to rotate when picked up under gravity. While a simple heuristic to prevent objects from twisting out of our grasp is to minimize the distance to the object’s centre of mass, here we present a study in which we demonstrate that human participants have much more nuanced sensorimotor understanding of how torque can be counteracted at our fingertips. Participants grasped L-shaped objects at various orientations while we tracked their fingertip movements. Stimuli were light-weight (wood, 70g) or heavy (wood/brass composite, 550g), designed to create grips with varying levels of torque. Depending on the object's orientation, participants either aligned their fingers to counteract torque through rotational friction or offset their fingertips to generate effective force couples (p<.001). These behaviours were further modulated by object mass, with heavier objects eliciting larger offsets (p<.001). These patterns were observable before the hand reached the object, demonstrating that these behaviours were visually-driven and predictive. Our findings thus suggest that humans leverage complex visually-based predictions of physical behaviours when selecting how to manipulate objects. Given recent methodological advancements, we are now poised to explore these processes for more complex, multi-digit actions in both natural and mixed-reality environments, deepening our understanding of the sophisticated interplay between perception, cognition, and motor control in grasping and object manipulation.