Ilmu Komputer & AI editorial
Distributed Trajectory Planning and Resource Allocation for Dynamic Multi-UAV Collaborative Computing
The core problem
The paper investigates a **multi-UAV-enabled distributed mobile edge computing (MEC)** framework in which the set of collaborating UAVs **changes dynamically over time**, driven by each UAV's energy state and service load. This dynamism breaks the assumptions of static-fleet optimization: UAVs may join, leave, or become unavailable mid-mission, so trajectory planning and resource allocation must be re-coordinated continuously and in a distributed manner.
The central problem is the **joint optimization of trajectory planning and resource allocation**. The authors formulate it as a **Stackelberg game** with two tiers:
- **Leaders (UAVs):** maximize their own benefit by balancing (i) executed workload, (ii) energy cost, and (iii) revenue from resource allocation.
- **Followers (mobile terminals, MTs):** minimize total overhead, composed of computing delay and resource cost, by choosing offloading ratios and resource requests.
The stated goal is a hierarchical, distributed solution that remains stable and efficient as UAV participation varies, rather than a centralized optimizer that assumes a fixed fleet.
Innovation
Simulation results demonstrate **stable convergence under dynamic UAV participation**, i.e., the learning-based coordination remains well-behaved even as the set of active UAVs changes over time.
Key quantitative findings reported by the authors:
- **+18.58% UAV efficiency** compared with the **no-collaboration benchmark**, achieved through **inter-UAV task migration**.
- **−33.77% average MT overhead** compared with the **fully offloading scheme**.
- The proposed algorithm **outperforms other benchmarks** under **varying network scales and capabilities**, indicating robustness to fleet size and heterogeneous UAV/MT resources.
The efficiency gain is attributed to jointly optimizing UAV operations (trajectory, migration, association, pricing) and resource utilization, rather than treating them as independent subproblems.
Why it matters
The results support three broader claims about dynamic multi-UAV MEC.
**1. Collaboration is the dominant lever.** The 18.58% efficiency improvement over the no-collaboration baseline comes specifically from **inter-UAV task migration**, meaning that letting UAVs hand off workload to better-positioned or less-loaded peers is more valuable than optimizing each UAV in isolation. This directly addresses the dynamic-participation setting: when a UAV's energy or load makes it a poor server, migration recovers the loss.
**2. Hierarchical game structure matches the incentive problem.** Modeling UAVs as leaders and MTs as followers captures the fact that UAVs set prices and association while MTs respond with offloading and resource requests. The two-stage iterative follower solution keeps MT-side computation lightweight, while MAPPO with CTDE gives UAVs a practical route to distributed execution without a central controller at run time.
**3. Robustness across scales matters for deployment.** Outperformance under varying network scales and capabilities suggests the approach is not tuned to a single fleet size or homogeneous hardware profile — an important property for real deployments where UAV availability fluctuates.
**Limitations and open directions.** The abstract reports convergence stability and benchmark comparisons but does not detail convergence-rate bounds, training sample complexity, or the sensitivity of the 18.58%/33.77% figures to specific energy models and pricing ranges. Future work could examine safety constraints, communication outages among UAVs, and scaling MAPPO to larger fleets.
Who should read this
Opening member content…