Potential Field Method (Swarm)

Agents navigate by performing gradient descent on the total potential field. The force F acting on an agent is the negative gradient of the potential: F(x) = -∇U_total(x). This force vector dictates the agent's acceleration and, consequently, its motion direction. The agent effectively "rolls downhill" on the potential surface toward the goal while being pushed away from high-potential obstacle regions. This provides a continuous, smooth trajectory.

A significant limitation is the creation of local minima—points where the sum of attractive and repulsive gradients is zero, trapping an agent. This occurs in symmetric obstacle configurations (e.g., a U-shaped obstacle or narrow corridor). The agent perceives a local low point but cannot reach the global goal. Mitigation strategies include:

• Adding random noise or virtual vortices to escape minima.
• Implementing navigation functions (guarantee a single minimum).
• Switching to alternative planners (e.g., A*) when stuck.

In tight spaces, an agent can experience oscillatory behavior or get stuck due to competing force vectors from closely spaced obstacles. As the agent moves, the dominant repulsive force can rapidly switch from one obstacle to another, causing jitter or back-and-forth motion. This challenges smooth transit and increases energy consumption. Solutions involve smoothing the potential field, dynamically adjusting gain parameters (k_rep, k_att), or using harmonic potential fields which are divergence-free.

The method is widely used for real-time swarm tasks due to its computational simplicity. Key applications include:

• Formation Control: Assigning different goal potentials to achieve geometric patterns.
• Coverage & Dispersion: Using inter-agent repulsion to spread out over an area.
• Dynamic Obstacle Avoidance: Repulsive fields are generated from moving obstacles detected via onboard sensors.
• Multi-Target Search: Attractive fields can be dynamically assigned to different targets for task allocation. It is foundational in Unmanned Aerial Vehicle (UAV) swarms and Autonomous Mobile Robot (AMR) fleets for warehouse logistics.

The Boid model is a seminal computational model for simulating flocking behavior using three simple, localized steering rules applied to each agent (a 'boid').

• Core Rules: Separation (steer to avoid crowding), Alignment (steer toward average heading of neighbors), and Cohesion (steer toward average position of neighbors).
• Relationship to Potential Fields: The Boid model is a rule-based, vector-addition approach, whereas Potential Fields use scalar gradient descent. Both achieve emergent coordination without central planning.

The Velocity Obstacle method is a decentralized, geometric approach for real-time collision avoidance among multiple moving agents.

• Mechanism: Each agent calculates a set of velocities in velocity space that would lead to a future collision with another agent or obstacle, then selects a velocity outside this set.
• Comparison: Unlike Potential Fields, which use force-based attraction/repulsion, VO is a velocity-space planning method that guarantees collision-free motion within a finite time horizon under specific assumptions.

Swarm path planning is the collective generation of efficient, collision-free trajectories for a large group of agents in a shared environment.

• Decentralized Focus: Each agent typically plans based on local sensor data and limited communication.
• Techniques Include: Potential Field Method, Velocity Obstacles, and rule-based models.
• Core Challenge: Balancing agent objectives (e.g., reaching goals) with swarm constraints (e.g., maintaining cohesion, avoiding collisions).

An Artificial Potential Field is the scalar function that defines attractive forces toward goals and repulsive forces away from obstacles. It is the core mathematical construct of the Potential Field Method.

• Attractive Field: Often shaped like a conic or parabolic well, pulling the agent toward the goal.
• Repulsive Field: Typically a high-value 'hill' or barrier around obstacles, pushing the agent away.
• Agent Motion: The agent acts as a particle moving downhill along the negative gradient of the combined field.