Due to their similarity to humans, humanoids are among the most promising type of general-purpose robots for the applications in natural environments such as homes, where robots need to coexist with people. To be effective in such environments, humanoids must become able to not only execute a number of preprogrammed operations, but also to acquire new knowledge and to react to unexpected situations in an open-ended way. This level of competence requires that humanoid robots are endowed with human-like learning capabilities, among them with efficient ways to acquire new sensorimotor knowledge.
Researchers in humanoid robotics have been aware for a long time that complex sensorimotor skills can most likely only be acquired through learning methods. Otherwise it is too complicated to solve problems arising from high-dimensional and highly nonlinear perception-action spaces. Various approaches have been explored in the past to solve different problems arising in sensorimotor learning on humanoid robots:
-
(1)model-based approaches, where an expert specifies physical models with various parameters that need to be learned to accomplish the task,
-
(2)emergence of sensorimotor knowledge through developmental progression, where motor skills are acquired through the dynamic interaction with the environment,
-
(3)statistical learning methods that can identify the relevant dimensions for learning directly from the data,
-
(4)imitation learning, where the learning process is initialized by transferring the demonstrator's performance to the humanoid, thus effectively limiting the search space, and
-
(5)learning from practice that gives the robot an opportunity to discover details that may have been missed during the observation.
The objective of this workshop is to gather researchers contributing to the field of sensorimotor learning on humanoid robots. By presenting the latest research results on various types of learning at one workshop, we hope to elucidate what kind of approaches are best suited to acquire new sensorimotor knowledge in different situations.