Humanoid Robots For Personal Assistance
Among various robotic manipulator systems in manufacturing and service mar-ket, the humanoid mobile manipulator systems have attracted increasing at-tention due to their human-like elegant appearance, dexterous moving and manipulation abilities. They can be classified into two categories: legged hu-manoid robot and humanoid robot with a mobile platform, according to the type of their lower body. The former would mainly play its role in non-flat situations like stairs, while the latter proves its conveniences in flat situations. The humanoid mobile manipulator systems can operate in an extended space due to the mobility of their lower body. They can mimic human beings with human accepted motions due to the upper human-like arm-hand manipulator system with dexterous end-eﬀectors like grippers or multi-fingered hands. In the past decades, many humanoid robots have been introduced (see Fig. 1.2).
The two-legged (biped) humanoid robot AMIO (2006)  in Fig. 1.2 (a) is created by KAIST A.I. & Media Lab1, Korea, for personal accompany. It has one head, two arms, two 3-fingered-hands and two legs.
The humanoid robot Atlas (2013)  in Fig. 1.2 (b) designed by Boston Dynamics company2, USA, is also a biped robotic system with two arms and a torso to achieve whole-body mobile manipulation, which greatly expands its workspace. The 130cm tall biped humanoid robotic system ASIMO (2010) in Fig. 1.2 (c) is designed by Honda3, Japan. It has the ability to recognize moving objects, pos-tures, gestures, its surrounding environment, sounds and faces, which enables it to interact with humans. However, its manipulation abilities have not been well-developed. The annual international robotics competition RoboCup  which is founded to promote robotics and AI research attracts many researchers each year. A group of biped humanoid robots are designed to form a soccer team and cooperate SoftBank Robotics (formerly Aldebaran Robotics), Japan, designed three hu-manoid robots: NAO (2006) , Romeo (2012)  and Pepper (2014) (see Figs. 1.2 (e), (f) and (g))4. In particular, NAO is a 58cm tall biped humanoid robotic system with 14-25-DoFs. However, it is designed for entertainment accompany, which can only oﬀer simple assistance. The 140cm tall biped humanoid robotic system Romeo is a research platform for assisting elderly people and people who experience mobility disabilities. However, it is only a mechanical base for research purpose and its manipulability ability remains to be explored. One of the challenges for designing and controlling such a biped robotic system comes from the physical instability and redundancy introduced by two legs. Thus, the humanoid robot with one mobile platform attracts people’s attention. The 120cm tall humanoid robotic system Pepper is such a robot which is designed for day-to-day accompany. It has one mobile platform, two arms and one communication screen, and is capable of identifying user’s emotion. However, it is only able to entertain and accompany people with embedded multimedia systems and to execute tasks with limited manipulability abilities.
On-line Motion Planning
Autonomous navigation and manipulator abilities of service robots are increas-ingly demanding. But most of the above mentioned methods mainly focus on the motion planning in static environments. The oﬀ-line planned motion will be no longer realizable and applicable when the environments are time-varying. Therefore, many works have been done on motion planning [57, 58, 100–123] in real time for robotic systems.
Primatesta  presented a trajectory planning algorithm based on Informed-RRT* for mobile robot navigation in crowded environments with moving people. Instead of computing the entire path in real time, it only evaluates the validity of the current path and partially repairs the path, thus often resulting in non-optimal paths. A RRT*-based motion planning algorithm consisting of reconnect and regrow processes was introduced in  for mobile robots in dynamic environments. However, the RRT* method which needs to sample throughout the task space between the initial and goal positions is time-consuming for long distance planning. Another disadvantage is that the random sampling process will generate diﬀerent paths for the same task. It treats the mobile robots as a point. Meng  introduced a BiRRT algorithm with pruning and re-planning processes of the global sampling tree to design paths for the unmanned aerial vehicle (UAV). Choi  proposed a two-level obstacle avoidance algorithm by introducing a multi-phase problem for an UAV system to avoid multiple obstacles with the minimal eﬀort. The motion planning method proposed by Greiﬀ  needs to find two feasible paths around an inflated obstacle by finding the projection points around the obstacles. And the process will become complex in a crowded and dynamic environment. Mercy  presented an optimization-based motion planning method based on separating hyper plane theorem and spline-based technique to generate motions for the non-holonomic vehicles. Mcleod  used Bezier curve to connect two via-segments of gener-ated path in real time for mobile vehicles. However, for non-holonomic mobile platform there will be too many orientation changes. On the other hand, in a crowded environment, spline-based trajectory generation may cause unnecessary obstacle collision.
We can see that all the above techniques are for collision-free motion plan-ning for mobile vehicles or free-flying robots. Although some of them can be applied to robots with high DoFs, their eﬀectiveness will decrease, and they are incompetent to solve multiple constraints problems. In order to solve this problem for robots, Lee  established an algorithm that provides an on-line trajectory generation for a dual-arm robotic system by introducing a virtual roadmap. Berg  proposed an algorithm to solve on-line motion planning for any robot type in configuration space with any dimension based on a roadmap. The roadmap is constructed for the static part of the environment, then only the dynamic part is dealt with on-line. A two-level search is performed to design the shortest path. On the local level, trajectories on single edges of the roadmap are found using a depth-first search on an implicit grid in state-time space. On the global level, these local trajectories are coordinated using an A*-search to find a global trajectory to the goal configuration. Pajak  proposed a motion planning method for a mobile surgery assistant. Only the knowledge of the end-eﬀector’s path is needed, and the reference trajectory of the mobile platform is not needed. It is based on a penalty function approach and a redundancy resolution at the acceleration level. The motion of the mobile manipulator is planned in order to maximize the manipulability measure, thus to avoid manip-ulator singularities. Akli  proposed a random profile approach (RPA)-based motion planning strategy for a mobile manipulator RobuTER/ULM to approach the target point by implementing the manipulability measure. Yang  intro-duced a configuration re-planning or adaptation method for a fixed manipulator in a dynamic environment. However, it requires Jacobian computation between the relative distance space and the joint space and it demands that the desired configuration must be known in real time.
In order to avoid dynamic obstacles during manipulation, Chen  de-signed an on-line joint velocity correction term to avoid obstacles by adjusting the manipulator’s joint trajectory. Xin  proposed a path planning algo-rithm for redundant robot arms in dynamic environments by introducing an escape velocity and projecting it onto the null space of Jacobian. An obstacle avoidance gain which depends on one minimum distance is defined to reduce computation burden in such a way to modify the IK homogeneous solution. The on-line obstacle avoidance method in  used the distance calculation and discrete detection for robot arms, but it is diﬃcult to be applied to humanoid mobile manipulators due to the demand of the obstacle identification and the dependence on robot’s structural complexity.
Nguyen  presented a heuristic planning method for a multi-DoFs hu-manoid robot’s arm by employing the sampling-based RRT* algorithm directly in task space in hierarchical fashion. However, it only considers the end-eﬀector and elbow as two hierarchical control points. In fact, as the number of control points increases, the hierarchical motion planning method will become much more complicated. Vannoy  introduced an on-line adaptive motion plan-ning algorithm by generating a population of trajectories based on the fitness evaluation for the mobile manipulators.
Motion Tracking and Controllers
The traditional PID controller still plays an important role in the control of robot systems in cooperation with the newly brought methods, such as relative motion control , force-position control , impedance control , task-priority method , etc. In addition, many advanced methods based on sliding mode , Fuzzy logic, Neural Network (NN), Chaos, and Expert experience have been introduced . Besides, in order to take advantages of multiple control techniques, some hybrid controllers  have been also investigated, like adaptive fuzzy control , robust adaptive fuzzy control , and robust adaptive NN control , etc.
Table of contents :
Table of Contents
List of Tables
List of Figures x
1 STATE OF THE ART AND PRELIMINARIES
1.1 STATE OF THE ART ON SERVICE ROBOTS
1.1.1 Developments and Applications of Service Robots
1.1.2 Humanoid Robots For Personal Assistance
1.2 LITERATURE SURVEY
1.2.1 Robotic System Modeling
1.2.2 Motion Planning Methods
1.2.3 Motion Tracking and Controllers
1.3.1 Forward Kinematics Background
1.3.2 Kane’s Method for Dynamic Modeling
1.3.3 Kinematic Control of Non-holonomic Mobile Platform
1.3.4 Non-dominated Sorting Scheme
2 SYSTEM DESCRIPTION AND MODELING
2.2 SYSTEM MODELING FOR MDAMS
2.2.1 System Description and Kinematic Modeling
2.2.2 Lagrangian Formulations and Control
2.2.3 Dynamic Modeling and Control using Kane’s Method .
3 MOTION PLANNING IN STATIC ENVIRONMENTS
3.2 MAXIMIN-BASED NSGA-II MOTION PLANNING
3.2.1 Improved MaxiMin NSGA-II Algorithm for Pose Design .
3.2.2 Motion Planning in Static Environments
3.3 SIMULATION RESULTS
3.3.1 Task 1: Chair A
3.3.2 Task 2: Chair D
3.3.3 Motion Planning and Dynamic Tracking for MDAMS
4 MOTION PLANNING IN DYNAMIC ENVIRONMENTS
4.2 MOTION PLANNING AMONG DYNAMIC OBSTACLES
4.2.1 On-line Motion Planning Algorithm
4.2.2 Via-points and MOGA-based Motion Planning
4.3 SIMULATION RESULTS
4.3.1 On-line Motion Planning for Mobile Platform
4.3.2 On-line Motion Planning for EE
4.3.3 Motion Planning Time-consuming Statistics
4.3.4 On-line Motion Planning for MDAMS
5 MOTION TRACKING FOR MDAMS
5.2 IK SOLVING AND JOINT LIMIT AVOIDANCE
5.2.1 IK Solving
5.2.2 Joint Limit Avoidance – mWLN
5.3 DUAL-TRAJECTORY TRACKING
5.3.1 Task Priority and Relative Motion-based IK Solving
5.3.2 Semi-decentralized Dual-trajectory Tracking
5.3.3 Simulation Results
5.4 TRIPLE-TRAJECTORY TRACKING
5.4.1 Task Priority and Motion Distribution-based IK Solving
5.4.2 Semi-decentralized Triple-trajectory Tracking
5.4.3 Simulation Results
CONCLUSIONS, CONTRIBUTIONS AND FUTUREWORK
A.1 Partial Velocity
A.1.1 Partial Velocity Table Construction
A.1.2 Vectors Zi (i = 1; :::;75) and Ai (i = 9; :::;75)
A.2 Matrix D 2 Rnn
A.3 Matrix B 2 Rn1