Rapid Prototyping Tools for Biped Robot “AIR”

Robot development process is complex and often very difficult to manage particularly when development processes are interconnected. It is not only a matter of determining physical features, but, beyond that, design, ergonomic, “time-to-market” and other demands must be taken into consideration as well. Consequently, „Rapid Prototyping (RP)” is the keyword not only in industry, but also in robotic research. A variety of RP processes are available allowing the implementation of special features. The paper presents an approach for designing and developing a biped robot for research purposes. 1 SYSTEM DESIGN AND OPTIMISATION BY VIRTUAL PROTOTYPING In practice, when designing mutually influencing subsystems of a robot system we proceed in the following order: mechanical structure, actuators, control systems (hardware) and control algorithm (software). The subsystems are often variable in development phase. In this case it will be helpful to use rapid prototyping tools. Nowadays there is a variety of tools providing good methods for development, designing and testing of mechanical, electronical or control structures. We tested one of these rapid prototyping concepts by developing the humanoid robot “AIR” presented in fig.1. FEM and rigid body dynamic simulation tools were used to define actuators, sensors and construction parameters. Fig. 1 Result of a biped mechanical structure RP To manage the variety of sensors for measurement of robot behavior, to control all of the actuators in real-time operation and to maintain the modular design of the robot a networked control system was developed. The high level control algorithms will be implemented in an external computer, which works on line and communicates with the robot over wireless interface. This provides flexibility of the control algorithm development and speeds up the development process. 2 MECHANICAL SYSTEM DESIGN AND OPTIMISATION Mechanical design and optimization strategies of a small-size humanoid walking machine are presented in this section. The main design goal is to keep the mechanical construction simple, lightweight and cheap. The design concept uses identical off-the-shelf high performance DC servo motors, linked together using simple sheet metal constructions. The servo motor with the parameters shown in table 1 was chosen because of the high torque to weight ratio. The humanoid robot “AIR” consists of total 25 degrees of freedom with 12 on the legs and waist and 11 on the upper body, 2 servo motors on the head to control orientation of the camera. To control these actuators and to process sensor data, a system of fast processing units is developed. The leg assembly has been modeled using 3D CAD software in order to provide features as functional as possible. Leg design and optimization process is shown on figure 2. Fig. 2 leg design and optimization The kinematics diagram, the three dimensional CAD model and the real prototype of the humanoid robot “AIR” are presented in figure 3. Numerical simulation and optimization tools for modeling the non-linear dynamic models are used throughout the entire design and development process to optimize kinematics and dynamic model parameters. Actuator placement was also optimized to obtain improved functionality of robot motion. The developed robot has 25 DOF. There are 2 DOF in a head, 6 DOF in each leg, 3 DOF in waist and 4 DOF in each arm. The mass of the robot is less than 1 kg without the batteries. The height equals 320 mm. Table 1 Specification of servo motor Speed: 40 grad/ 0,09s Dimension: 28.5 x 28.5 x 13 [mm] Torque: 55 [Ncm] Mass: 19 [g] Supply Voltage: 6V Fig. 3 Kinematical diagram, 3D model and real prototype Sensor system Three single-axis gyroscopes and 2 dual-axis accelerometers are installed in the robot trunk for sensing acceleration and rotation in reference frame axes. Four single-axis force sensors are placed in each robot foot to determine dynamic stability by mean of ZMP estimation. Additional accelerometers are mounted on the legs and arms for sensing purposes. Each servo motor module consists of potentiometer for joint absolute angle measurement and current sensor. These sensors provide appropriate position and force control of the robot joints. Additionally, a thermal sensor detects overheating of each servo module. [3] 3 REAL-TIME (RT) SIMULATION AND CONTROL TOOLS Recently to humanoid robots additional features such as vision processing, environment sensing, adaptive locomotion are required. Thus, software for sensor signal processing and for control of humanoid robots will become a nontrivial task as more functionality is expected. In addition, it is desirable that any programs created for the robots could be easily ported to new generations of hardware. Robot’s structure, such as the internal control electronics or communication system, should be of no interest. Obviously, primary software (the higher level cognitive functions and gait generation) should operate regardless of hardware changes [4]. In our work a RT-Simulator Tool concept and implementation is presented. It provides device independence through the hardware abstraction layer (HAL). The HAL is a hardware-specific interface, provided by prototype hardware developer, which uses the RT-Simulator to work directly with the prototype robot hardware. Control applications never interact with the HAL directly i. e. the infrastructure of the HAL provides the RT-Simulator engine with a consistent set of interfaces and methods that an application uses to control the robot behavior. The HAL can be a part of the prototype firmware or a separate dynamic-link library (DLL) that communicates with the prototype firmware over a transport layer. The HAL implements only device-dependent code and performs no emulation. If a function is not performed by the hardware, the HAL does not report it as a hardware capability. Foot, 2 DOF Neck, 2 DOF Hip, 3 DOF Waist, 3 DOF Arm, 4 DOF The RT-Simulator engine provides three different processing modes on the same robot prototype hardware: • software control processing; • hardware control processing; • hybrid control processing. The software control processing mode is a pure software control, which can be used even without the robot prototype hardware for simulation purposes, i.e. this mode allows software development for special applications to design the robot control scheme, even if the robot hardware does not exist yet or is characterized by unknown behavior. The hardware control processing mode is a pure hardware control, which can be used for real-time simulation and control of the robot prototype hardware. This mode can be used on the existing robot hardware. The hybrid control processing mode is a combination of hardware and software control, which can be used for real-time simulation and real-time control of robot prototype hardware without some parts (i.e. sensors or actuators). This mode can be used on the existing part of robot hardware, and works in real-time simulation with the hardware mapped on the model. The behavior of the part that does not exist will be simulated in the same control system (software in the loop). In such way, for example, the variety of sensors can be simulated or virtual sensors implemented in the real hardware. The structure of the RT-Simulator engine and corresponding robot side client is shown in figure 4. Fig. 4 Real-time simulator and control system architecture The drive abstraction layer was developed to make control algorithms independent of the actuator type and regulator architecture. Upper abstraction levels will not depend on e.g. the type and number of regulators in this layer. Robot-side client PC-side server client GUI & Visualisation, Data Managing User input devices