Careful with That! Observation of Human Movements to Estimate Objects Properties
This work addresses the challenge of improving human-robot interaction by allowing robots to interpret subtle movement cues, though it is incremental as it builds on existing machine learning methods for visual analysis.
The study tackled the problem of enabling robots to infer object properties from human movements, successfully demonstrating that machine learning can reliably detect carefulness in object handling from visual data, but found the same approach inadequate for distinguishing between light and heavy objects.
Humans are very effective at interpreting subtle properties of the partner's movement and use this skill to promote smooth interactions. Therefore, robotic platforms that support human partners in daily activities should acquire similar abilities. In this work we focused on the features of human motor actions that communicate insights on the weight of an object and the carefulness required in its manipulation. Our final goal is to enable a robot to autonomously infer the degree of care required in object handling and to discriminate whether the item is light or heavy, just by observing a human manipulation. This preliminary study represents a promising step towards the implementation of those abilities on a robot observing the scene with its camera. Indeed, we succeeded in demonstrating that it is possible to reliably deduct if the human operator is careful when handling an object, through machine learning algorithms relying on the stream of visual acquisition from either a robot camera or from a motion capture system. On the other hand, we observed that the same approach is inadequate to discriminate between light and heavy objects.