Abstract
Imagine a friend hands you a plain cardboard box. You would probably want to know whether it contained bricks or feathers before trying to grab it. Understanding the physical properties of objects is critical in order to interact with the world effectively. Unfamiliar objects have latent properties, such as their weight, that are not directly observable. In previous work, we showed that observers can accurately determine which of two objects is heavier by seeing another person lift them. Furthermore, weight judgements were lifter-invariant – comparisons of objects lifted by two different people were just as accurate as comparisons of objects lifted by the same person. What visual cues support lifter-invariant weight perception? Candidate cues should be consistent across lifters without the need to be normalized to lifter-specific behaviors. Here, we sought to characterize kinematic features of object lifts and test which, if any, could be directly compared across different lifters to infer object weight. We filmed four individuals each lifting ten visually indistinguishable objects that ranged in weight from 100g – 2000g. In a series of behavioral experiments, we verified that: i) observers could precisely discriminate objects’ weights after viewing lifts, ii) the precision of discrimination was lifter-invariant (not diminished when making comparisons across different lifters), and iii) performance was consistently good throughout the task, suggesting that cues were already known, not being learned on-the-fly. Next, we built models to discriminate object weights using annotated kinematic features from the videos: peak lifting and lowering speeds and overall lift durations. Comparing any one of these features to make trial-by-trial decisions about relative weight yielded performance on par with behavioral experiments and a close match to participants’ responses. These results suggest that simple kinematic features are robust across different lifters and sufficient to account for observers’ performance in weight lifting observation tasks.