Introduction
In 1992, researchers at JPL challenged the prevailing paradigm of autonomous robotics. At the time, it was widely believed that navigating complex, unstructured environments required significant computational power to build internal world models and plan intricate paths. However, for planetary exploration, such requirements are prohibitive due to power and mass constraints. This paper presents Rocky III, a small rover that demonstrates how "intelligence" can emerge from the synergy of clever mechanical design and a reactive, behavior-based control architecture.
The Challenge of Planetary Navigation
Planetary rovers face a unique set of constraints. Large, power-hungry computers are liabilities on space missions. Furthermore, the high-bandwidth communication necessary for remote tele-operation is often unavailable. The central question addressed by this research was whether a "computationally impoverished" robot—equipped with only an 8-bit processor and 10KB of memory—could perform meaningful scientific tasks autonomously in rugged terrain.
The Rocky III System: Hardware and Software Synergy
The success of Rocky III relied on two core components: mechanical intelligence and a decentralized control framework.
Mechanical Intelligence: The Rocker-Bogie Suspension
Instead of relying on complex sensors to avoid every small pebble, Rocky III utilizes a rocker-bogie suspension. This springless system allows the rover to passively traverse obstacles larger than its own wheels. By handling minor terrain irregularities mechanically, the burden on the software is significantly reduced.
Behavior-Based Control (ALFA)
The software architecture, known as ALFA (A Language For Algorithms), eschews a central planning "brain." Instead, it employs a hierarchy of independent behaviors, such as "Follow Heading" and "Move Around Obstacles," which are tightly coupled to sensor inputs. Higher-level tasks provide goals to these behaviors without micromanaging their execution.
The sensor suite was intentionally minimalist, featuring a flux gate compass, pitch/roll clinometers, wheel encoders, and simple contact switches for obstacle detection.
Experimental Results
The system was rigorously tested in both indoor laboratories and the outdoor Arroyo Seco wash near JPL, which features sand, loose rocks, and large boulders intended to simulate a rugged planetary surface.
- Extreme Efficiency: The entire control system operated on a single 8-bit Motorola 6811 processor using just 10KB of memory.
- Autonomy and Robustness: In numerous trials, Rocky III successfully navigated to waypoints, performed soil sampling (using a specialized scoop mechanism), and returned to a lander beacon. Failures were almost exclusively hardware-related rather than logic errors.
- Reactive Navigation: The "bump and retreat" strategy proved effective for natural terrain. By treating the environment as its own model, the rover navigated hazardous areas despite its limited sensory input.
Discussion and Future Work
While Rocky III proved the power of reactive navigation, the authors noted certain limitations. The "blind" obstacle avoidance—relying on physical contact—could be improved with non-contact sensors. Additionally, while effective in natural environments, purely reactive systems might struggle in artificial "maze-like" traps where a robot could get stuck in a local minimum. Future work points toward hybrid systems that integrate reactive control with symbolic reasoning for even higher mission complexity.
Reference: Miller, D. P., Desai, R. S., Gat, E., Ivlev, R., & Loch, J. (1992). Reactive Navigation through Rough Terrain: Experimental Results. AAAI-92 Proceedings.
Summary updated