Autonomous Guidance and Control of a Roving Robot

J. A. MillerIJCAI 1977

Abstract

In the late 1970s, NASA and the Jet Propulsion Laboratory (JPL) sought to transition from strictly teleoperated machines to semi-autonomous rovers for planetary exploration. This paper details the development of an early autonomous guidance system that replaces inefficient "stop-turn-go" navigation with a "steering-while-moving" strategy. By utilizing symmetric maneuver planning and a hierarchical real-time software architecture, the system balances traverse speed with path accuracy, providing a foundation for modern robotic locomotion.

Introduction

A major hurdle in early planetary exploration was operational efficiency. Standard navigation sequences at the time required robots to come to a complete stop before rotating to a new heading, then restarting motion. While mathematically precise, this method is slow, energy-inefficient, and places high mechanical stress on steering hardware.

The challenge of steering while in motion is that it creates a variable-radius turn, causing the robot to deviate from its intended path. This research addresses how to execute these maneuvers smoothly without requiring constant re-planning from high-level navigation modules, effectively managing the trade-off between speed and transverse error.

System Architecture and Methods

The researchers implemented a "Guidance System" as a core component of a larger locomotion software suite. The technical approach focused on three primary areas: maneuver geometry, software hierarchy, and real-time control.

Symmetric Maneuver Planning

To minimize path deviation during a turn, the system utilizes a symmetric planning strategy. Steering is initiated at a calculated distance before a prescribed turning point and concluded at an equal distance after it. This approach "centers" the error along the intended path, theoretically eliminating transverse (sideways) error by the time the turn is completed.

Hierarchical Software Design

The software is split into a multi-tiered architecture to manage computational load and hardware responsiveness:

  • Guidance Module: A high-level module running on a 1-second cycle that manages the overall path.
  • MOVE and STEER Programs: Low-level, high-priority programs that handle immediate motor control.
  • Real-Time Operating System (RTOS): To prevent jerky movement or motor stalls, the authors employed a clocked interrupt system. This ensured that the hardware received precise control signals even when the main guidance logic was processing complex calculations.

The hardware utilized a gyrocompass and optical wheel encoders to maintain a "task-relative Rover state." Because the main guidance loop operated with a half-second delay, the software was designed with predictive state management. This allowed the robot to "know" its current trajectory while simultaneously preparing the control parameters for the next execution cycle.

Experimental Setup

Evaluation was performed in a controlled laboratory environment featuring a flat surface and constant illumination. As this was an engineering-focused study on guidance laws rather than a machine learning application, the "dataset" consisted of physical performance metrics gathered during live traverses. The robot relied on dead-reckoning to determine its position relative to the planned route.

Key Insights

  • The Speed-Accuracy Trade-off: The authors identified vehicle speed as the primary constraint on maneuver design. Higher speeds increase the theoretical magnitude of deviation during turns, requiring the guidance system to dynamically limit speed based on the maximum allowable transverse error.
  • Subsystem Autonomy: A core philosophy of the design was that the guidance system should handle its own error correction. By fixing local drifts internally, the system avoided taxing the high-level Path Planning Module for new routes unless absolutely necessary.

Results and Performance

The system was successfully implemented in both hardware and software, demonstrating that "steering-while-moving" is a viable alternative to stop-and-go methods. Key performance metrics included:

  • Longitudinal Error: Approximately 1% of the total distance traveled.
  • Transverse Error: Less than 10% of the turn arc length (achieved even before active transverse correction logic was fully optimized).
  • Orientation Error: Final heading errors were maintained at approximately 1 degree.

Limitations and Future Work

While the results were promising, the authors noted several areas for improvement. The specific software module for correcting transverse errors was incomplete at the time of publication; its completion was expected to reduce sideways drift by a factor of five. Additionally, the bulk of the remaining error was attributed to steering hardware limitations rather than algorithmic failures.

Finally, the reliance on dead-reckoning in a controlled lab environment remains a challenge for field deployment. Future work must adapt these guidance laws to rugged planetary terrain where wheel slippage would significantly degrade the accuracy of encoder-based navigation.

Summary updated

BibTeX
@inproceedings{miller1977autonomous,
  author = {J. A. Miller},
  title = {{Autonomous Guidance and Control of a Roving Robot}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {1977},
}
Figure