When people interact with objects, they show incredible flexibility in learning novel motor control mappings or adapting their known control mappings to variables like object mass. Such motor learning can benefit from intuitive physical reasoning, as novel contexts of object interaction could be a new combination of a previously experienced control mapping with a different object with known mass. In this work, we present a novel object interaction paradigm in which subjects learned to slide pucks at targets by releasing kinetic energy from a compressed spring in a computer game. Participants needed to learn how their motor actions related to the final positions of the puck, while also adapting to the mass of different pucks. With a Bayesian regression model, we inferred participants' beliefs about object mass and control mappings, and show that they could transfer information about previously experienced puck mass but not the motor mappings of the springs.