Modern robotics applications are becoming more complex due to greater numbers of sensors and actuators. The control of such systems may require multiple processors to meet the computational demands and to support the physical topology of the sensors and actuators. A distributed real-time system is needed to perform the required communication and processing while meeting application-specified timing constraints. We are designing and implementing a real-time kernel for distributed robotics applications. The kernel's salient features are consistent, user-definable scheduling, explicit dynamic timing constraints, and a two-tiered interrupt approach. The kernel wi1l be evaluated by implementing a two-arm robot control example. Its goal is to locate and manipulate cylindrical objects with spillable contents. Using the application and the kernel, we will investigate the effects of time granularity, network type and protocol, and the handling of external events using interrupts versus polling. Our research will enhance understanding of real-time kernels for distributed robotics control.