A Mobile Robot with Onboard Parallel Processor and Large Workspace Arm

Rodney A. Brooks, Jon Connell, Anita M. FlynnAAAI 1986

Introduction

In the mid-1980s, mobile robotics faced a significant challenge known as the "power/weight spiral." To achieve intelligent behavior, robots required substantial computing power, which necessitated heavy batteries and larger motors, further increasing weight and power demands. Consequently, most robots of the era were either tethered to stationary mainframes or moved with extreme slowness. In 1986, Rodney Brooks and his team at the MIT AI Lab introduced MOBOT-2, a robot designed to overcome these limitations through a decentralized control philosophy and an onboard parallel processor.

The MOBOT-2 mobile robot platform.

The Computational Bottleneck

Traditional robotic systems typically relied on "functional decomposition," a vertical pipeline where data passed through discrete stages: perception, modeling, planning, and execution. This architecture created a single point of failure; if any stage processed data too slowly, the entire system stalled. Furthermore, the reliance on offboard Lisp machines for high-level reasoning introduced communication bottlenecks that hindered real-time interaction. MOBOT-2 addressed these issues by integrating all processing onboard using a robust, behavior-based approach.

Methodology: Subsumption Architecture and BARNACLE

The core innovation of MOBOT-2 lies in its "subsumption architecture." Rather than vertical functional slices, the system is organized into horizontal, task-achieving layers.

The horizontal decomposition of the subsumption architecture, showing layers of task-achieving behaviors.

In this model, lower-level behaviors (such as "avoid objects") operate independently and can "subsume" higher-level goals (such as "wander" or "explore") when necessary. For instance, if the robot is wandering but detects an obstacle, the avoidance layer suppresses the wander command to steer the robot safely. This ensures inherent robustness; even if a complex high-level module fails, the robot maintains basic survival behaviors.

To implement this architecture, the researchers developed BARNACLE, a distributed system comprising 32 Hitachi 6301 microcontrollers. Unlike conventional computers with a central bus, BARNACLE utilizes a fixed physical topology. The "brain" is literally wired using a patch panel of subminiature phone jacks, where suppression and inhibition are handled by dedicated hardware timers and flip-flops rather than software overhead.

Key Insights and System Design

The design of MOBOT-2 emphasizes simplicity and efficiency:

  • Fixed Topology: By hard-wiring communication paths, the authors eliminated the need for complex, power-hungry switching networks.
  • Data Representation: The system uses 24-bit packets (3 bytes), which the authors identified as the "sweet spot" for representing spatial coordinates $(x, y, \theta)$ or motion commands with sufficient precision.
  • Optimized Manipulation: The robot features a 2-degree-of-freedom (DOF) planar arm optimized for a massive vertical workspace (40 inches). The researchers argued that lateral dexterity can be compensated for by moving the robot base, allowing for a lighter, more efficient arm design.

Results and Performance

MOBOT-2 successfully demonstrated full onboard autonomy, a significant advancement over its predecessor, MOBOT-1. The BARNACLE system achieved a polling rate of 500 microseconds per port, enabling 24-bit packet transmission in approximately 0.1 seconds—performance levels that surpassed previous offboard simulations. Despite its lightweight construction, the arm achieved 1/4 inch precision across its vertical range. Additionally, the robot's hand was equipped with infrared range-finders, providing local topographical data to assist in grasping tasks.

Conclusion and Future Work

While the 2-DOF arm and manual reconfiguration of the BARNACLE patch panel present certain limitations in dexterity and adaptability, MOBOT-2 proved that complex, real-time robotic behavior could be achieved without massive centralized computing. This work laid the foundation for behavior-based robotics, moving away from the "power/weight spiral" toward smaller, more agile autonomous systems.

Summary updated

BibTeX
@inproceedings{brooks1986mobile,
  author = {Rodney A. Brooks and Jon Connell and Anita M. Flynn},
  title = {{A Mobile Robot with Onboard Parallel Processor and Large Workspace Arm}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1986},
}
Figure