Technical Contributions. In order to allow the development and dissemination of oneM2M Technical Specifications and Technical Reports, it is necessary that each Partner Type 2 and oneM2M member contributing to the technical work of oneM2M grant a perpetual, worldwide, royalty-free, non- exclusive license to: (i) incorporate text, graphics or other material whether oral or written from the contributed material in oneM2M Technical Specifications and Technical Reports; and
Technical Contributions. From and after the Termination Date, TJA and TSI shall each remain fully vested with all right, title and interest (as owner, licensee or designee, as the case may be) in and to its respective Technical Contributions (as defined in Section 1.14 of the Original Agreement).
Technical Contributions. Following the prototype deliverable findings, we have addressed the challenges common to using IRL techniques in robotics such as poor scalability, lack of efficient representation that admit robotic constraints and lack of incremental reward learning using iterative IRL algorithms. Our contributions are summarized as follows; • We developed Controller Graphs (CG) which are an efficient graph based representation for reinforcement learning in large and possibly constrained domains. CGs then allow us to apply IRL to large realistic robot tasks while incorporating tasks constrains such as non-holonomy common in mobile robotics, (see [Okal and Arras, 2016] and [Okal et al., 2015] for more de- tails, including attached paper). • We develop a new iterative IRL algorithm (TBIRL) based on the Bayesian IRL formulation that requires only a generative model of the underlying MDP, hence eliminating the need for full state and action space enumeration. Furthermore, our algorithms works on sampled trajectories to evaluate candidate rewards making it applicable to very large (possibly infinite) domains.
