A Vision-based Scheme for Kinematic Model Construction of Re-configurable Modular Robots

摘要

Re-configurable modular robotic (RMR) systems are advantageous for theirreconfigurability and versatility. A new modular robot can be built for aspecific task by using modules as building blocks. However, constructing akinematic model for a newly conceived robot requires significant work. Due tothe finite size of module-types, models of all module-types can be builtindividually and stored in a database beforehand. With this priori knowledge,the model construction process can be automated by detecting the modules andtheir corresponding interconnections. Previous literature proposed theoreticalframeworks for constructing kinematic models of modular robots, assuming thatsuch information was known a priori. While well-devised mechanisms and built-insensors can be employed to detect these parameters automatically, theysignificantly complicate the module design and thus are expensive. In thispaper, we propose a vision-based method to identify kinematic chains andautomatically construct robot models for modular robots. Each module is affixedwith augmented reality (AR) tags that are encoded with unique IDs. An image ofa modular robot is taken and the detected modules are recognized by querying adatabase that maintains all module information. The poses of detected modulesare used to compute: (i) the connection between modules and (ii) joint anglesof joint-modules. Finally, the robot serial-link chain is identified and thekinematic model constructed and visualized. Our experimental results validatethe effectiveness of our approach. While implementation with only our RMR isshown, our method can be applied to other RMRs where self-identification is notpossible.