شرح حرکت ربات و مدیریت زمان واقعی با پروتکل توضیحات حرکت هارمونیک

We describe the Harmonic Motion Description Protocol (HMDP), that can serve as a part in tools for rapid prototyping of behaviors in a hybrid simulation real robot environment. In particular, we are focusing on the RoboCup 3D Soccer Simulation League server that is currently evolving rapidly, and becomes a more and more useful open source, general purpose simulation environment for physical Multiagent systems. HMDP uses harmonic functions to describe motions. It allows for superposition of motions and is therefore capable of describing parametric motions. Thus, typical open loop motions (walking on spot, forward, turning, standing up) of small humanoid robots are readily available and can be optimized by hand. In conjunction with the HMDP some software tools and a small real-time motion generator (called Motion Machine) have been developed. The current implementation is very flexible to use and can easily be implemented in rather small embedded systems.

Many developers of autonomous robot systems experience difficulties when designing a control system that is at the same time capable of high level sensor processing–in particular vision sensors–and motor control. This is particularly true for humanoid robots that can have around 20 degrees of freedom. For advanced systems (e.g. Honda’s Asimo or Sony’s Qrio robots) multitasking real-time OS systems (often with several CPUs) are available, that manage the entire sensor and actuator processing in real time. Low cost designs usually lack the sensor processing and are restricted to a remote controlled embedded CPU or even an analogue system (e.g. RoboSapiens, [1]). These systems completely lack autonomous behavior capabilities. For the medium to low cost designs (for example in RoboCup [2] and [3], where autonomous behaviors are demanded), the solution can be a hybrid design using 2 CPUs, one for motor control and one for sensor processing. On the one hand the sensor data processing–in particular vision–is done by a portable IBM-PC i386 like system with a custom broadband multitasking operating system (Windows, Linux, BSD) that does not have real-time capabilities (see for example [4], [5], [6] and [7]). The drivers for cameras and other devices are cheap and do not need further development. The motor control on the other hand is done by a micro-controller that performs pre-defined motion patterns. The demanded motion pattern is communicated between both CPUs by a serial pipe or bus system, sometimes wireless. In particular, in humanoid robots, the motion control part has to be a real-time system. Motion patterns such as standing up need to be precisely executed in the range of a time-span of 100 ms, in order to produce a reliable performance. For this reason direct positional control from the PC side should be avoided.

We presented here an environment for the design and management of motions in small, medium priced humanoid robots. We assume for our system that the robots use the hybrid architecture of a non-real-time CPU for sensor processing and a real-time motion controller. Between these two CPUs we suggest a standardized protocol that can also be used in robot simulations for the communication between agent programs and the simulator. In addition, we presented a tool for motion design using the proposed protocol. The advantage of the described method to previous designs is that in comparison to streaming of robot posture information the communication load can be reduced. At the same time motion design and management are still highly flexible (all kinds of parameter motions are possible). One important disadvantage to a standard control program on the motor controller is that closed loop approaches are more complicated to implement. With regard to the outlook we focus in this section on two points: How closed loop control can be applied and on the role HMDP can serve in Simspark (RoboCup 3D soccer simulation environment. It is planned to add the code there). Fig. 9 illustrates both subjects. In the case of the closed loop control using HMDP one may design a compensation motion pattern for a specific expected perturbation (in the figure: healing pattern). At a specific phase of the motion pattern the Motion Machine can trigger a sensor reading. In the case of alarming sensor values the healing pattern can be activated in an appropriate amplitude. In this way although the feedback delay may be relatively large, the reaction precisely fits into the motion pattern.