The paper describes a significant development in the intelligent automatic control of an adaptive robot. The particular application concerns the problem of autonomous excavation, and the subtlety of the approach lies in the fact that the machine can cope with highly variable ground conditions and even underground obstructions without human interference. The nature of the excavation problem is first described. The development of a working fifth-scale model of an excavator—LUCIE—the Lancaster University Computerised Intelligent Excavator. The hardware systems are briefly described, but the main emphasis of the paper concerns the method by which the AI technique of a rule-based “production system” is used to control the tip of the excavator bucket. The software is in two distinct parts; a low-level velocity controller and a high level “activities manager” which contains the rules. It is concluded that the best way to achieve subtle, human-like, automatic control is by means of simple rules and heuristics, and not complex control algorithms. Finally the paper reports field trials of LUCIE, and goes on to describe proposed future developments.