AML: A general human-inspired motion language for anthropomorphic arms based on movement primitives.

The recent increasing demands on accomplishing complicated manipulation tasks necessitate the development of effective task-motion planning techniques. To help understand robot movement intention and avoid causing unease or discomfort to nearby humans toward safe human–robot interaction when these tasks are performed in the vicinity of humans by those robot arms that resemble an anthropomorphic arrangement, a dedicated and unified anthropomorphism-aware task-motion planning framework for anthropomorphic arms is at a premium. A general human-inspired four-level Anthropomorphic Arm Motion Language (A2ML) is therefore proposed for the first time to serve as this framework. First, six hypotheses/rules of human arm motion are extracted from the literature in neurophysiological field, which form the basis and guidelines for the design of A2ML. Inspired by these rules, a library of movement primitives and related motion grammar are designed to build the complete motion language. The movement primitives in the library are designed from two different but associated representation spaces of arm configuration: Cartesian-posture-swivel-angle space and human arm triangle space. Since these two spaces can be always recognized for all the anthropomorphic arms, the designed movement primitives and consequent motion language possess favorable generality. Decomposition techniques described by the A2ML grammar are proposed to decompose complicated tasks into movement primitives. Furthermore, a quadratic programming based method and a sampling based method serve as powerful interfaces for transforming the decomposed tasks expressed in A2ML to the specific joint trajectories of different arms. Finally, the generality and advantages of the proposed motion language are validated by extensive simulations and experiments on two different anthropomorphic arms.