⚡ electrical engineering

Obstacle Avoidance of UAV in Dynamic Environments Using Direction and Velocity-Adaptive Artificial Potential Field

This paper proposes a novel Direction and Relative Velocity Weighted Artificial Potential Field integrated with Model Predictive Control to enable UAVs to effectively navigate dynamic environments by resolving local minima and generating smooth, collision-free trajectories that account for obstacle kinematics.

Original authors: Nikita Vaibhav Pavle, Rakesh Kumar Sahoo, Manoranjan Sinha

Published 2026-08-25

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Nikita Vaibhav Pavle, Rakesh Kumar Sahoo, Manoranjan Sinha

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a drone flying through a busy city sky, tasked with delivering a package or inspecting a bridge. For this machine to operate safely, it must do more than just follow a pre-drawn line on a map; it must react instantly to moving cars, other aircraft, and sudden gusts of wind. This is the challenge of autonomous navigation: teaching a machine to see the world, predict where things are going, and make split-second decisions to avoid a crash. For decades, engineers have relied on a method called the artificial potential field to solve this. In this approach, the drone's destination acts like a magnet, pulling it forward, while obstacles act like invisible walls that push it away. The drone simply follows the path of least resistance, sliding between the pull of the goal and the push of the danger. However, this simple system has a fatal flaw. If the pull from the goal and the push from an obstacle happen to balance each other out perfectly, the drone can get stuck in a dead zone, hovering helplessly in mid-air, unable to move forward or backward. This is known as the local minima problem, and it becomes even more dangerous when the obstacles themselves are moving, as the static rules of the old system cannot predict where a moving object will be next.

Researchers at the Indian Institute of Technology Kharagpur have proposed a new way to fix this, designed specifically for drones flying in dynamic, cluttered environments. Instead of treating every obstacle as a static wall, their new system gives the drone a sense of direction and speed. The team developed a method that weighs the danger of an obstacle based on two specific factors: where the obstacle is relative to the drone's path, and how fast the obstacle is moving toward or away from the drone. If an obstacle is directly in front of the drone, the system treats it as a severe threat and pushes the drone away strongly. If the obstacle is behind the drone, the push is much weaker, allowing the drone to move forward without being unnecessarily repelled. Furthermore, the system calculates the relative speed between the drone and the obstacle. If a fast-moving object is approaching from the side or behind, the system recognizes the high-speed threat and increases the repulsive force to create a larger safety buffer, prompting the drone to steer clear earlier and more smoothly.

To put this new logic into action, the researchers combined it with a sophisticated planning tool called Model Predictive Control. Think of this as a system that constantly looks a few seconds into the future, simulating different paths the drone could take to find the smoothest, safest route that obeys the laws of physics. The team tested their new approach in computer simulations against the traditional method. In one test, they created a scenario designed to trap the old system in a dead zone between obstacles. The traditional drone hovered and stalled, unable to break free. In contrast, the drone using the new direction and velocity-aware system successfully navigated around the cluster, finding a clear path without getting stuck. In a second, longer test involving a cluttered environment with moving objects, the new method proved superior in safety and efficiency. The traditional drone often waited until it was dangerously close to an obstacle, sometimes dropping to a mere two meters away, before making a sharp, reactive turn. The new system, however, began its avoidance maneuvers much earlier. By anticipating the speed and direction of the moving obstacles, it maintained a wider safety margin and executed smoother, more predictable turns.

The results of these simulations suggest that adding an awareness of direction and relative speed to the drone's navigation logic significantly improves its ability to handle complex, real-world scenarios. The new method does not just react to what is immediately in front of the drone; it understands the flow of traffic around it. This allows the drone to resolve the stuck-in-place problem that has plagued older systems and to fly with a level of caution that feels more like a human pilot than a simple robot. While these findings are currently limited to computer simulations, the researchers indicate that the next step involves testing the system on physical drones to see how it handles the noise and delays of real-world sensors. If successful, this approach could make autonomous flight safer and more reliable, paving the way for drones to operate confidently in the busy skies of the future.

Technical Summary: Obstacle Avoidance of UAV in Dynamic Environments Using Direction and Velocity-Adaptive Artificial Potential Field

Problem Statement
The paper addresses the critical challenge of autonomous collision avoidance for Unmanned Aerial Vehicles (UAVs) operating in dynamic and cluttered airspace. While local reactive methods like the classic Artificial Potential Field (APF) are favored for their low computational cost, they suffer from two fundamental limitations:

Local Minima: The equilibrium between attractive (goal) and repulsive (obstacle) forces can trap the UAV, preventing it from reaching its destination.
Lack of Dynamic Awareness: Conventional APF treats obstacles as static, failing to account for the kinematics of moving obstacles, specifically their direction and relative velocity. This can lead to unsafe, reactive maneuvers rather than predictive avoidance.

Methodology
The authors propose a novel Direction and Relative Velocity Weighted Artificial Potential Field (APF) integrated within a Model Predictive Control (MPC) framework.

System Modeling: The UAV is modeled as a quadcopter using a simplified second-order integrator (double integrator) for the outer control loop, relating position and velocity to acceleration inputs. This continuous-time model is discretized for MPC implementation.
Control Architecture: The MPC framework solves a constrained quadratic programming (QP) problem over a prediction horizon. It minimizes a cost function that balances tracking error (deviation from the reference path generated by APF) and control effort, subject to kinematic constraints (e.g., maximum speed and acceleration).
Proposed APF Formulation:
- Directional Weighting: A weighting function $\omega(\theta)$ is introduced based on the angle $\theta$ between the UAV's velocity vector and the relative position vector to the obstacle. This increases the repulsive force for obstacles directly in the path of motion while reducing it for obstacles behind the UAV.
- Relative Velocity Weighting: To address dynamic threats, a second term is added based on the relative velocity ( $v_e$ ) between the UAV and the obstacle. This term scales the repulsion based on whether the obstacle is approaching or receding.
- Combined Weight: The final repulsive potential is scaled by a bounded function $\omega(\theta, v_e) = [1 + \gamma(1 + \cos\theta)/2] \cdot (2 + k \tanh(G))$ , where $G$ represents the relative velocity component along the line of sight. This ensures the repulsive force is dynamically adjusted: it is strongest when an obstacle is ahead and approaching rapidly, and weaker when the obstacle is behind or receding.

Key Contributions

Novel Weighting Function: The introduction of a bounded weighting function that dynamically scales the repulsive potential based on both the relative direction and relative velocity of obstacles.
Integration with MPC: The formulation of this adaptive APF as a reference generator within an MPC framework, ensuring that the resulting trajectories adhere to the UAV's kinematic constraints.
Resolution of Local Minima: The method provides a mechanism to escape local minima by preventing the complete neutralization of forces through directional biasing.

Simulation Results
The proposed method was evaluated against a conventional Basic APF in two scenarios:

Static Local Minima Test: In a configuration designed to trap a standard APF, the Basic APF caused the UAV to stall. The proposed Velocity-Weighted APF successfully navigated around the obstacle cluster, demonstrating superior path integrity.
Dynamic/General Path Following: Over a 300-second simulation in a cluttered environment:
- Safety: The proposed method maintained a smoother approach to obstacles and initiated avoidance maneuvers earlier compared to the Basic APF, which exhibited sharp drops to the minimum safe clearance distance.
- Efficiency: The Velocity-Weighted APF demonstrated a consistent rate of distance reduction to the goal, avoiding the oscillations and stalls seen in the Basic APF.
- Trajectory Quality: 3D visualizations confirmed the generation of stable, collision-free paths with smooth transitions in a cluttered 3D space.

Significance and Claims
The paper claims that integrating direction and velocity awareness into the APF framework provides a reliable and computationally efficient solution for autonomous path planning. The primary significance lies in enhancing safety margins by enabling predictive avoidance maneuvers rather than purely reactive ones. By resolving the local minima issue and accounting for dynamic threats, the system ensures superior path integrity, making UAV operation more feasible in complex, dynamic airspace.

Future Work
The authors outline several avenues for future research, including:

Real-Time Implementation: Validating the system on physical quadrotor platforms to assess computational overhead and latency.
Uncertainty Handling: Incorporating sensor noise and measurement uncertainty into the MPC objective function.
Multi-Agent Coordination: Extending the approach to multi-UAV systems for cooperative navigation.
Energy Optimization: Modifying the MPC cost function to balance safety with energy consumption.

More like this