Biped Locomotion Control In Indoor Environment

More and more researchers paid attention to humanoid robots due to its outstand-ing humanoid shape. Currently, humanid robots are limited in indoor environment. Although this kind of environment is better than out-door conditions, biped locomo-tion control is still a challenge problem due to its intrinsic non-linaear property, high-dimension property, and instability. In this dissertation, after introducing the history of humanoid robots, a detailed research status will be given in terms of "Stability Crite-rion","Gait Pattern Generation", and "Feedback Control".The word "Flexibility" comes from RoboCup. Robots that have flexible locomo-tion always have higher ball controlling percentage. In the future, humanoid robots will provide services to human beings and flexibility will be an important property in various application scenarios. First of all humanoid robots are required to respond to high level commands as soon as possible. Secondly, they need the ability of avoiding obstacles in a fast manner. Last but not the least, higher speed must be achieved to finish the task in shorter time. In this dissertation, a method of generating flexible gait pattern will be given. This method will not treat the ZMP of one step as constant point but use two cubic splines for sagittal direction and lateral direction. ZMP trajectories and CoG tra-jectories of every control cycle will be simultaneously planned. The maximum delay of responding to high level command is one step duration, while the minimum one is one control cycle. This method enables the robot change to arbitrary footstep in one step while playing dynamic walking. The traditional problem of connecting discrete ZMPs in double support phase is naturally avoided which plays an important role in increas-ing walking frequency. Besides, for the problem of uncontrollable ZMP trajectories, an optimization method and a refine method are given. The optimization method takes the goal of minimizing double support phase, so that the robot will have more time in single support phase. The refine method is performed in the case that the ZMP trajectories are out of the support polygon.A new physical model will be introduced in this dissertation. This model can repre-sent the angular momentum of the trunk of the robot which is important for increasing flexibility once more. The detailed deduction of its dynamics will be given and the discrete numerical solution of the resulting dynamics will be discussed too.Recently, some high-level robots are developed and showed fancy locomotion such as running, climing stairs, walking on uneven terrain that is not modeled in advance. And of course, this fancy locomotion gained profit from its high level hardware which makes those robots really expensive. In the year2008, robot NAO become the standard robot of standard platform league of RoboCup. NAO has a lot of sensors and a much cheaper price, so that it is accepted in many laboratories. This kind of low-level robots are conquering the world showing that humanoid robots will no longer be facilities of laboratory but furnitures of family. However, they suffer from low level sensors and elastic joints which lead to motor deflections. This dissertation will do massive researches on NAO.First of all, the method of feed-forward compensation for elastic joint is given. It is based on Lagrange principle and obtains the compensation equations via inverted pendulum model. Besides, some optimization techniques are introduced to accelerate the computation and make it be able to be computed in real time.Strategies integrated for different application scenarios are discussed in this disser-tation. When walking on planar surface, two strategies are used. They are closed-loop gait pattern generator based on inertial sensors and ankle feedback technique. When walking on uneven terrain, four strategies are integrated. Two of them are closed-loop gait pattern generator based on joint sensors and posture feedback control based on in-ertial sensors. The rest two strategies, i.e. ground reaction force control and CoG height control are used in emergency situation. This method enables the robot to walk on un-even terrain which is not modeled in advance. The uneven terrain consists of obstacles and a free-move slope.The main contributions of this dissertation are showed as follows.1. A method that satisfies flexibility benchmarks is given. Footstep is generated according to the latest high level command. The maximum delay of responding is one step duration, while the minimum delay is one control cycle. This satisfies the first benchmark which is "Responding to high level command in time". The robot is able to change to arbitrary footstep using just one step without any preview time. This sat-isfies the second benchmark which is "Change footstep in real time". This dissertation implemented a fast walking of which step duration is just0.18[s]. The delay of respond-ing is just0.18[s]. The maximum forward speed is0.33[m/s]. This satisfies the third benchmark which is "Reach target point as soon as possible"2. A compensation method completely based on theory is discussed while the tra-ditional compensation is performed in an empirical way. It is based on the theory of Lagrange function. Linear inverted pendulum is also used to compute the compensa- tion. Several techniques are used to optimize the computation time resulting a real time method. Moreover, a technique of computing elastic coefficient is derived.3. Feedback strategies are integrated deliberately for planar surface and uneven terrain. When walking on planar surface, the robot is able to perform stable high fre-quency walking. When walking on uneven terrain, the robot is able to walk on various obstacles and a movable slope. This work enables the robot to perform fancy locomo-tion just like the high level robots.4. A new physics model is presented. The detailed deduction of its dynamics based on angular momentum theorem and D’Alembert principle is given. Moreover, a numer-ical solution of its dynamic equation is provided and its truncation error is discussed in detail. This model can represent angular momentum of the trunk of the robot which further enhances the flexibility of biped walking.