Real-time obstacle avoidance algorithm for unmanned vehicle systems — Chapter 2: Implementation

drone in agriculture

In the second and final part of this article series, we are going to discuss the implementation of a new hybrid algorithm for autonomous obstacle avoidance and path planning.

Access to full research paper: https://arxiv.org/abs/2506.20311

part 1 of the article series: https://medium.com/@vishwa.gw1998/real-time-obstacle-avoidance-algorithm-for-unmanned-vehicle-systems-chapter-1-f560777ca3c5

Safety navigation for mobile robots has become a crucial challenge, especially in unfamiliar or unpredictable terrain. The lack of prior information regarding the environments poses major difficulties in predicting navigation routes, such as the unpredictability of the movements of obstacles on unknown terrains and the ambiguity of their shapes.

The global path planning has two primary categories. They are search-based methods and sampling-based methods. Traditionally, algorithms such as A, Dijkstra, D, etc. Recent advancements have focused on optimizing these algorithms by refining heuristic functions and addressing concerns such as node redundancy and selection strategies.

Reactive navigation can generate the path planning process with collision avoidance in a dynamic or unknown environment. So far, Model Predictive Control (MPC) has been widely recognized as an effective reactive approach capable of identifying the optimal trajectory by leveraging predictive models of motion.

Focus area:

During this research study, scientists identified the problem first. In this study, they were focusing on building a 3d forest environment populated by a series of obstacles, both static and dynamic. The goal is to develop a navigation law that allows the robot to move safely within this environment, avoiding the spreading fire and reaching its destination.

Hybrid navigation:

This proposed safety navigation system contains 3 major layers. They are the observation layer, the execution layer and the decision layer.

• The observation layer:

The observation layer of the proposed safety navigation approach is primarily composed of the global path planning algorithm. The researchers have used the RRT algorithm as it is very time-efficient. RRT is a sampling-based search algorithm designed to swiftly identify collision-free paths in a certain environment. It is a sampling-based search algorithm that is being designed to swiftly identify collision-free paths in a certain environment.

• Decision Layer:

The proposed system will have two operation modes. One is offline mode, which is a routine operation mode, and the other is online mode, which is a local/reactive escape operation mode.

In this layer, the distance to the closest object/obstacle acts as the sole external criterion for the decision-making process. There is also an additional time-observer in the online mode for more accurate obstacle avoidance.

• execution layer:

This layer also contains several modes.

The path tracking mode is mainly when the UAV is following its optimal path until an obstacle is encountered in the path. The distance to the unknown obstacle becomes the sole determining factor during the operation. Then the reactive mode is activated.

The Reactive mode is mainly when the UAV is required to avoid an obstacle. Inspired by biological systems, the researchers employed a reactive control system based on local motion control principles.

This approach utilizes distance information to enable collision-free motion around moving objects by identifying the smallest turning angle required for avoidance. The distance detected by the sensor is the sole necessary information to implement this method, enabling local collision-free navigation in dynamic situations.

Final notes on the Hybrid algorithm:

• This research presented a navigation algorithm combining a global planner (RRT) and a reactive planner for the mobile robot’s safe navigation in a simulated forest fire scenario. The global pre-defined path generated by RRT is applied in the execution layer.
• To handle local obstacles and ensure a quick response, a boundary-following reactive method is employed.
• The proposed navigation algorithm’s performance is evaluated through MATLAB simulations, using specific cases tailored to the forest fire scenario. These simulations serve to validate the effectiveness and efficiency of the algorithm in navigating through the simulated environment.