Abstract
This paper considers a two-dimensional persistent monitoring problem by controlling movements of second-order agents to minimize some uncertainty metric associated with targets in a dynamic environment. In contrast to common sensing models depending only on the distance from a target, we introduce an active sensing model which considers the distance between an agent and a target, as well as the agent's velocity. We propose an objective function which can achieve a collision-free agent trajectory by penalizing all possible collisions. Applying structural properties of the optimal control derived from the Hamiltonian analysis, we limit agent trajectories to a simpler parametric form under a family of 2D curves depending on the problem setting, e.g. ellipses and Fourier trajectories. Our collision-free trajectories are optimized through an event-driven Infinitesimal Perturbation Analysis (IPA) and gradient descent method. Although the solution is generally locally optimal, this method is computationally efficient and offers an alternative to other traditional time-driven methods. Finally, simulation examples are provided to demonstrate our proposed results.
Introduction
Persistent monitoring involves robots moving through an environment to keep "uncertainty" low at specific target locations. Most existing research simplifies this by assuming robots can change velocity instantly (first-order dynamics) or by ignoring physical obstacles and robot size. In reality, robots have mass and limited acceleration, and their sensors often perform worse at high speeds (e.g., motion blur in cameras). Furthermore, real-world mission spaces are cluttered with obstacles. Solving the "optimal" path for multiple robots under these constraints usually leads to a Two-Point Boundary Value Problem (TPBVP) that is mathematically elegant but computationally intensive. This paper seeks a middle ground: trajectories that are physically realistic, collision-free, and computationally tractable.
Methodology
The proposed approach breaks the problem down into several key components:
- Second-Order Dynamics: Agents are controlled via acceleration ($u_n$) rather than direct velocity, with explicit bounds on both acceleration and speed.
- Active Sensing Model: Unlike standard models that only consider distance, this model assumes sensing effectiveness drops as the agent's velocity increases. This creates a trade-off: agents want to move fast to cover more ground but must slow down near targets to "see" them clearly.
- Collision Penalization: The objective function includes "penalty" terms for agent-to-agent and agent-to-obstacle collisions. Obstacles are modeled as circumscribed circles for simplicity.
- Parametric Trajectories: To avoid the "curse of dimensionality," agent paths are restricted to specific families of 2D curves, specifically ellipses and Fourier series trajectories.
- IPA-Based Optimization: Infinitesimal Perturbation Analysis (IPA) is used to calculate gradients. IPA is "event-driven," meaning it only performs calculations when specific events occur (e.g., an agent enters a target's sensing range or hits a collision boundary). This is significantly more efficient than traditional time-driven gradient descent.
Experimental Results
The algorithm was evaluated in a simulated $10 \times 5$ unit mission space with 66 target points and circular obstacles.
In all simulations, the algorithm successfully converged to trajectories where the distance to obstacles and other agents remained above the safety threshold. As shown in the figure above, Agent 1 follows an elliptical path that avoids the yellow obstacle regions while covering the target grid.
In a single-agent elliptical case, the performance metric $J$ (representing total uncertainty) dropped from an initial high-cost state (with collisions) to a stable local optimum of 662.6 after 36 iterations. Fourier series trajectories provided even more flexibility, achieving a slightly better uncertainty score of 654, though with higher trajectory complexity.
Conclusion
This research demonstrates that combining velocity-dependent sensing with event-driven optimization allows for the design of efficient, collision-free trajectories for second-order agents. While the current method finds local optima and is centralized, future work will explore decentralized solutions and strategies for handling dynamic events in real-time.
Summary updated