Complexity and Similarity Metrics for Unsupervised Identification of Programming Methods for Robot Cooking Tasks

Using robotics to automate complex manipulation tasks such as cooking could offer significant advantages. Robotics could offer improved throughput, provide assistance for those that lack the required mobility or allow for creative robotic chefs. Generating a robotic system that utilizes vision systems, planning and dexterous manipulation would offer a generalized solution to kitchen tasks. However, there can be many components of the overall task that do not require this generalization. In addition, there remain many tasks for which it is quicker and preferable for a human to complete some sub-components of the task. By computing a complexity and similarity metrics from multiple examples of a human teaching the required motion we propose using unsupervised clustering to determine the approach method of implementation. This includes hard coding, use of vision, play back or human completion. We apply this analysis to the task of cooking pancakes in a kitchen environment, analyzing the trajectories, determining the method of completing the task using unsupervised methods and implementing the proposed approach.