A VLSI robotics vector processor for real-time control

A VLSI robotics vector processor (RVP) for real-time control is described. Hardware parallelism and pipelining is used to exploit potential concurrency in the low-level matrix/vector operations characteristic of the kinematics and dynamics computations required for control. Three floating point processors (FPP), each with a adder, multiplier, and register file, all operating in parallel in an SIMD (single-instruction, multiple data stream) fashion are incorporated. Data exchange between the FPPs is facilitated by a dual shift-broadcast network. High-speed dual I/O channels are provided so that input/output bottlenecks are avoided. The RVP uses a RISC (reduced-instruction-set-computer) architecture with seven basic instructions. Composite vector operations such as matrix-vector multiply and vector cross product may be readily programmed using the basic instruction set to given considerable overlap. The RVP can be implemented on a single VLSI chip using 1.2-μm CMOS