[Paper Review] Solving a Path Planning Problem in a Partially Known Environment using a Swarm Algorithm

Esh Vckay, Mansimar Aneja+1 more|arXiv (Cornell University)|May 9, 2017

Robotic Path Planning Algorithms18 references3 citations

TL;DR

This paper proposes a hybrid path planning approach for autonomous ground vehicles in partially known outdoor environments using a biomimetic swarm algorithm inspired by termite nest building. The method combines global path planning via a rank-based terrain evaluation (using elevation and soil data) with real-time local obstacle avoidance using the Vector Field Histogram (VFH) technique, achieving robust navigation in simulated complex terrains with dynamic obstacles.

ABSTRACT

This paper proposes a path planning strategy for an Autonomous Ground Vehicle (AGV) navigating in a partially known environment. Global path planning is performed by first using a spatial database of the region to be traversed containing selected attributes such as height data and soil information from a suitable spatial database. The database is processed using a biomimetic swarm algorithm that is inspired by the nest building strategies followed by termites. Local path planning is performed online utilizing information regarding contingencies that affect the safe navigation of the AGV from various sensors. The simulation discussed has been implemented on the open source Player-Stage-Gazebo platform.

Motivation & Objective

• To address the computational complexity of path planning in partially known, dynamic outdoor terrains for non-holonomic robots.
• To reduce processing load by leveraging distributed swarm intelligence inspired by termite behavior.
• To integrate multi-source environmental data (elevation, soil type) into a unified terrain suitability metric for global path planning.
• To enable real-time local obstacle avoidance using sensor data and the VFH algorithm.
• To validate the hybrid approach in a simulated environment using the Player-Stage-Gazebo platform.

Proposed method

• Global path planning uses a spatial database containing digital elevation models (DEM) and soil information from GRASS GIS, processed via a biomimetic swarm algorithm inspired by termite nest building.
• A rank value is computed per grid cell as the sum of soil goodness (1–5) and gradient goodness (1–5), based on elevation differences and soil type, to assess terrain suitability.
• The swarm algorithm identifies navigable regions (nests) by simulating agent interactions that favor higher-ranked cells, forming an optimal path from start to destination.
• A dual-grid system with overlapping 1/4-cell offsets improves resolution and reduces path inefficiency due to obstacle representation.
• Local path planning uses the Vector Field Histogram (VFH) technique, processing laser sensor data to generate safe, real-time steering commands.
• Soil composition is assessed via a simulated digital cone penetrometer, and data is used to guide the robot toward more traversable terrain during local navigation.

Experimental results

Research questions

• RQ1 Can a biomimetic swarm algorithm effectively generate a globally optimal path in a partially known environment using terrain attributes?
• RQ2 How does combining elevation and soil data into a single rank metric improve path planning robustness?
• RQ3 To what extent does the dual-grid system reduce path inefficiency caused by coarse obstacle representation?
• RQ4 How effective is the VFH algorithm in maintaining safe local navigation when unexpected obstacles are introduced?
• RQ5 Can real-time soil sensing enhance local path planning in variable terrain conditions?

Key findings

• The swarm algorithm successfully generated a globally optimal path by identifying high-rank cells based on combined elevation and soil suitability.
• The dual-grid system with 1/4-cell offset reduced path inefficiency caused by coarse obstacle representation, improving navigability.
• The VFH algorithm effectively avoided newly introduced obstacles (simulated crates) not present in the initial grid, maintaining safe local navigation.
• The integration of real-time soil data from a simulated cone penetrometer enabled the robot to preferentially navigate toward more suitable terrain, such as gravel and sand.
• The simulation ran stably on an Intel Core 2 Duo system with 2GB RAM, demonstrating feasibility despite limited computational resources.
• Visual results in Gazebo showed clear distinction between the initial global path (solid line) and local path deviations (dotted line) due to obstacle avoidance.