Maximum Sustainable Yield Problem for Robot Foraging and Construction System

Ruohan Zhang, Zhao Song

Introduction

As robotics moves from controlled environments to complex ecosystems, the goal of foraging shifts from simple maximization to long-term sustainability. This paper addresses the challenge of harvesting regenerative resources—such as biological matter or timber—where aggressive extraction can lead to population collapse. The authors propose a framework based on the biological concept of Maximum Sustainable Yield (MSY): the peak rate at which a resource can be harvested without depleting its base population over time.

The Challenge of Unknown Environments

The primary difficulty in achieving MSY in the real world is that the mathematical parameters of resource growth are rarely known beforehand. Furthermore, robot sensors are inherently noisy. Traditional approaches might attempt complex curve-fitting, which is computationally expensive and data-intensive. This research seeks a self-regulating system that can "blindly" discover and maintain the MSY point using simple, robust heuristics.

Methodology: Adaptive Foraging and Construction

The system models resource growth using a standard Logistic Model. To manage harvesting without prior knowledge of growth parameters, the authors developed an Adaptive Algorithm. This algorithm leverages a key property of the logistic curve: growth rate peaks when the population is at exactly half of its carrying capacity.

The robots follow a specific state machine to manage their tasks and regulate harvesting pressure dynamically.

Robot state machine illustrating the cycle of harvesting, unloading, and the 'sleep' state used for regulation.

A critical component is the "Simultaneous Decrease" Heuristic. The system monitors both the resource population and its growth rate. When both metrics begin to decrease simultaneously, it signals that the harvesting pressure has pushed the population just past the MSY point. At this stage, the system stops adding new foragers. To fine-tune the harvest rate without decommissioning robots, the system adjusts a tsleep variable, effectively making robots wait longer between foraging trips to lower the overall pressure.

Results and Performance

The adaptive algorithm was tested in the Player/Stage simulator. Results demonstrated that the system could successfully converge to the MSY population level without any prior knowledge of the environment's growth model.

Experimental results showing (a) observed growth and (b) population size, highlighting the stop condition reached by the adaptive algorithm.

While the system is robust to moderate noise in robot actuation and resource growth (up to 10%), it is highly sensitive to observation noise. The study found that if errors in counting the resource population exceed 1.5%, the robots can no longer reliably detect the "simultaneous decrease" signal, leading to potential overharvesting or instability.

Conclusion

This work demonstrates that a swarm of robots can achieve ecological sustainability through simple, decentralized heuristics. By mimicking biological self-regulation, the system balances productivity with resource preservation. Future research will focus on improving robustness to observation noise through decentralized information sharing and exploring dynamic environments with seasonal growth fluctuations.

Summary updated

Abstract

We introduce the Maximum Sustainable Yield problem for a multi-robot foraging and construction system, inspired by the relationship between the natural resource growth and harvesting behaviors in an ecosystem. The resources spawn according to the logistic model and are vulnerable to over-harvesting. The robots must maintain sustainability while maximizing productivity. The foraging robots harvest different types of resources, which enable a construction robot to build new foraging robots. We design algorithms to perform robot construction, assignment, and scheduling. We propose an adaptive algorithm to overcome the problem that resource growth model is often unknown. We demonstrate that our algorithms are robust to noises in the actuation and the environment. The case where the observation noise could harm sustainability is discussed.
BibTeX
@inproceedings{zhang2016maximum,
  author = {Ruohan Zhang and Zhao Song},
  title = {{Maximum Sustainable Yield Problem for Robot Foraging and Construction System}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2016},
}

Loading related content...

Figure