Clementine: Colorado School of Mines

AAAI 1996

TL;DR

This paper introduces "Clementine," a hybrid deliberative/reactive robot architecture developed by undergraduate students at the Colorado School of Mines. The system aims to provide practical experience in robotics by integrating classroom learning with real-world challenges, focusing on robust navigation and task execution in complex environments.

What Problem Does This Paper Address?

The paper addresses the challenge of providing undergraduate students with a hands-on, integrated robotics experience that bridges theoretical knowledge with practical application. Specifically, it focuses on developing a robot capable of navigating and performing tasks in potentially hazardous indoor environments, such as those found in industrial settings or emergency response scenarios. The goal is to equip students with skills in complex robot control, sensor fusion, and real-time decision-making, preparing them for demanding applications in the field of mobile robotics.

Methods

The core of the system is the "Clementine" platform, a Denning-Branch MRV-4 research robot equipped with ultrasonics, a laser navigation system, and cameras. The robot's control architecture is a hybrid deliberative/reactive system. The deliberative layer handles high-level planning, receiving task goals and generating a path plan using a topological map. The reactive layer then executes this plan by controlling constituent behaviors.

Key innovations include:

  • Scripts for Coordination: Novel use of "scripts" (originally from Natural Language Processing) to coordinate concurrent and sequential activities within the reactive layer. These scripts act as templates for managing collections of behaviors over time.
  • Behavior Partitioning: Dividing behaviors into "strategic" and "tactical" categories. Strategic behaviors (e.g., NavigateHall) generate high-level directions, while tactical behaviors (e.g., avoid-obstacle, fuzzy speed-control) interpret these directions in the context of immediate sensor data to produce actual robot actions.
  • Sensor Modality Integration: The students are required to use multiple sensing modalities, specifically ultrasonics for obstacle avoidance and basic navigation, and computer vision for room identification, to gain practical experience.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Bridging Theory and Practice: The project successfully demonstrates how an undergraduate curriculum can be enhanced by providing a platform for students to apply theoretical concepts to a complex, real-world robotics problem.
  • Hybrid Architectures are Key: The paper highlights the effectiveness of hybrid deliberative/reactive architectures for robots operating in dynamic and uncertain environments, allowing for both high-level planning and low-level responsiveness.
  • Scripts as a Powerful Abstraction: The innovative use of scripts for behavior coordination offers a flexible and structured way to manage complex robot tasks, simplifying the development of sophisticated behaviors.
  • Strategic vs. Tactical Behavior Separation: This clear distinction in the reactive layer allows for more modular and understandable robot control, where high-level goals are translated into actionable commands based on real-time sensor feedback.

Key Results

The paper describes the system's objectives and architecture rather than presenting quantitative results from a competition or benchmark. The key outcome is the successful implementation of the Clementine robot and its hybrid architecture, enabling it to perform tasks such as navigating through halls and doors, avoiding obstacles, and identifying rooms. The students gained practical experience in integrating and modifying code for research projects.

Limitations and Future Work

The paper does not explicitly detail limitations or future work. However, it implies that the system is designed for indoor environments and specific tasks like surveillance and site assessment. Future work could involve extending the robot's capabilities to more complex outdoor environments (as hinted by the mention of the C2 robot) or to a wider range of hazardous scenarios. Further research could also explore more advanced sensor fusion techniques and more sophisticated behavior scripting.

Summary updated

BibTeX
@inproceedings{1996clementine,
  title = {{Clementine: Colorado School of Mines}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1996},
}
Figure