The Spinoza robot consists of an Real World Interface (RWI) B-12 base with 3 black and white video cameras for trinocular stereo. There are two TMS320C40 DSPs for image processing, two T805 transputers for control and a T225 transputer for communication. The robot can communicate to a Sun Ultra Sparc 1 workstation over high speed radio modems. One of the DSPs has two SGS-Thompson A110 convolver chips added to accelerate stereo processing. The DSPs were chosen because of their image processing power while the transputers were chosen for their ease of use in real time applications, communications, and distributed processing.
Figure 2 shows the data flow through the subsystems and across the hardware can be seen in Figure 2. The cameras are genlocked so that the first C40 can capture a field of video simultaneously from all three. This is transfered to the second C40 which computes a trinocular depth image. It then computes a horizontal planar map that contains the nearest obstacle in each direction the cameras can see. The planar depth map and the current location and heading of the robot (as computed from odometry) is sent via the router and radio modems to the host. On the host, the mapper program uses the planar map to update its occupancy grid representation of the environment. The environment map is passed on to the planner/explorer module which computes safe and desired paths for the robot. These paths are sent back down to the robot where the controller communicates with the RWI base to execute the path.
