| With the ongoing intensification of societal aging,there has been a gradual increase in the demand for elderly care services,resulting in a growing shortage of caregivers.Hu-manoid service robots（HSR）,characterized by their human-like appearance and advanced operational capabilities,have emerged as a promising solution to alleviate the societal bur-den of elderly care.However,the complexity of human-centric work environments and the diverse care service requirements pose significant technological challenges for the practical implementation of HSR.Specifically,in the field of motion generation,there are several unresolved issues in existing research.Firstly,there is a need for comprehensive and ef-ficient analytical inverse kinematics（IK）algorithms to improve the planning and control efficiency of service robots equipped with anthropomorphic manipulators with radial elbow offset（AMREO）.Secondly,the high degree of freedom of AMREO makes it challenging for end-users to provide convenient operation demonstrations,leading to a lack of high-quality demonstration data to support motion generation.Thirdly,existing kinematically optimal motion generation algorithms do not systematically consider the naturalness of robot mo-tions,resulting in generated motions that may appear weird.Additionally,in close-range operations,the random motions of humans often interfere with the existing robot motions,and conventional obstacle avoidance algorithms may not be able to effectively avoid whole-arm collisions in a timely manner,posing a safety risk to human-robot interactions.In light of these challenges,this thesis presents an in-depth investigation into the theo-ries and technologies pertaining to the motion generation of HSR.The primary focus of the research encompasses the following content,along with the innovative outcomes achieved:（1）The IK Solution Method for AMREO.To begin with,a kinematic model of AM-REO is established,and the redundancy of the manipulator is described using the arm angle parameterization method.In particular,the unique elbow configurations of“outward-offset”and“inward-offset”are analyzed,which distinguish AMREO from conventional redundant manipulators.By decoupling the degrees of freedom,a closed-form solution to the IK prob-lem is derived,allowing for efficient computation of joint angles from end-effector pose.The proposed solution algorithm exhibits remarkable computational efficiency,with an av-erage computation time of approximately 3 microseconds,which is two orders of magnitude faster than numerical iterative methods.Moreover,a significant improvement in solution efficiency,amounting to 37.4%compared to similar solvers,is achieved.（2）Real-time Motion Transfer Strategy from Human Arm to Robotic Arm.In order to mitigate the challenges associated with teaching humanoid robots and provide data support for motion generation algorithms,a real-time motion imitation method based on a vision-based motion capture system is proposed.The approach involves transforming the problem of motion imitation into a simultaneously tracking problem of dual-target trajectories for the wrist and elbow points.A task hierarchy of multiple imitation objectives is then constructed,and a layered optimal control strategy for trajectory tracking is established,taking into ac-count joint physical constraints.To efficiently solve this strategy,a hierarchical recurrent neural network,named HQP-SWET-RNN,is designed with guaranteed convergence and sta-bility.The proposed imitation control algorithm achieves a high level of tracking accuracy,with a comprehensive accuracy of～10-4m for dual-target trajectories,and a computation time of less than 1 ms per control loop,reducing computation time by 12.5%compared to conventional solving methods.Furthermore,this method demonstrates robustness against disturbances in arm length ratio and the ability to avoid joint limits.（3）Human-like Motion Generation Algorithm for AMREO.In order to enhance the naturalness of motion during the operation of humanoid robots,a motion generation algo-rithm based on natural arm configuration learning is proposed.Taking inspiration from the habitual arm configuration of human arm reaching movements,the natural arm configura-tion of AMREO is defined.A Natural Arm Angle Prediction Network（NAPN）based on a Long Short-Term Memory（LSTM）network is constructed,allowing AMREO to learn the implicit association between wrist positions and natural arm configurations from human motion demonstrations.For the motion generation problem,the arm angle is used as a key variable,which is transformed into a path-wise inverse kinematics problem.A redundant resolution framework that incorporates natural arm configurations is then constructed.Com-pared to traditional planning methods,the algorithm proposed in this thesis generates joint motions that exhibit smoother trajectories and higher levels of humanoid-like movements.（4）Hybrid Motion Replanning Method for AMREO.In order to mitigate the risk of col-lisions during close-range operations,a novel algorithm framework for motion replanning based on online motion modulation strategies is proposed.The framework utilizes an im-proved RRT-Connect path planning component to initialize and optimize a global collision-free path in joint space.Additionally,a hybrid local path replanning component is con-structed,which effectively identifies and replaces invalidated segments of the pre-planned path due to interference,without requiring a complete replanning of the entire path.Fur-thermore,a quintic polynomial-based online trajectory generation component is designed to generate joint control commands with continuous acceleration based on the dynamic path.The proposed method reduces the replanning time by 47.2%compared to conventional meth-ods,while increasing the success rate of obstacle avoidance by 35%.Based on the research findings discussed earlier,this thesis develops a human-centric prototype system for robot-assisted elderly care services.The proposed functional modules are integrated to create a cohesive system.A Smart Aging Room is established,and the functionality of the prototype system are rigorously tested through skill teaching and task instances involving item fetching and delivery.The experimental results validate the effec-tiveness of the proposed system in reducing caregiver workload and enhancing the quality of robot-assisted services. |
PDF Full Text Request