Autonomous Mobile Robots Get Smarter: New Algorithm Cuts Airport Baggage Handling Time by Over 11%
Logistics inside modern airports has long resembled a high-stakes ballet—complex, tightly choreographed, and unforgiving of missteps. Every second counts, especially when it comes to baggage handling: delays cascade into missed connections, frustrated passengers, and costly operational bottlenecks. Amid rising global air traffic—45.4 billion passengers and over 2 billion checked bags moved in 2019 alone—the pressure on ground systems is intensifying. Traditional baggage handling, reliant on fixed conveyor belts or magnetic-tape-guided AGVs (Automated Guided Vehicles), is increasingly seen as inflexible, expensive to reconfigure, and ill-suited for the dynamic demands of next-generation “smart” airports.
Enter the Autonomous Mobile Robot, or AMR—a new breed of intelligent, sensor-laden vehicle that navigates freely without embedded infrastructure, adapts in real time to changing environments, and promises unprecedented agility. Yet, as engineers and logistics planners are discovering, deploying a fleet of AMRs is far more challenging than running a single unit. The true bottleneck lies not in mobility, but in task assignment: how to efficiently match dozens of dynamically arriving baggage jobs—each with unique origin and destination points—to a swarm of robots whose positions, availability, and workloads are in constant flux.
A newly published study from researchers at Sichuan University and the Civil Aviation Administration of China’s Second Research Institute offers a breakthrough. Their work, titled “AMR Task Allocation Based on Improved Greedy Algorithm,” introduces a refined scheduling strategy that significantly trims both decision latency and total system runtime—without sacrificing robustness or scalability. In simulations mimicking the high-pressure environment of a check-in island at a major hub, their method outperformed established benchmarks by at least 8.4% in runtime, and as much as 13.7% under moderate traffic conditions. More impressively, it achieves these gains using only lightweight algorithmic enhancements, making it practical for real-world deployment on modest onboard computing hardware.
What makes this approach noteworthy isn’t raw computational firepower, but operational intelligence—a design philosophy that mirrors how seasoned human dispatchers think. Instead of brute-force optimization or complex auction-style negotiations, the team, led by Jin Xie, Yong Xiang, and Xiuqing Yang under the supervision of Professor Xin-Zhi Zhou, re-engineered the classic greedy algorithm, injecting three key insights derived from real airport constraints.
First, they replaced simplistic geometric distance estimates—like Euclidean or Manhattan metrics—with path-aware cost calculations using the A search algorithm. This may seem like a minor technical tweak, but its impact is profound. In cluttered terminal layouts where baggage carts, security barriers, and maintenance zones create impassable zones, straight-line distance is often meaningless. A robot parked 20 meters away might be inaccessible due to a wall, while one 30 meters distant enjoys a clear corridor. By precomputing realistic traversal costs—including obstacle avoidance and turn penalties—the system avoids costly misassignments that waste time and energy. In one illustrative test case included in the paper, Euclidean distance would have dispatched a robot blocked by a service kiosk; A-based routing correctly selected a slightly farther—but actually reachable—unit, saving an estimated 5.4 seconds per trip. Multiply that across hundreds of daily assignments, and the savings compound dramatically.
Second—and perhaps most critically—the team introduced state-aware prioritization of robots. Rather than treating all AMRs as interchangeable candidates every time a baggage item appears at a check-in counter, they categorize robots into three operational modes: (0) idle (stationed in the staging zone), (1) returning (heading back to staging after a drop-off), and (2) active (currently en route with a bag). Crucially, the system leverages a hierarchical selection policy: when a new task arrives, it first offers it only to idle robots; if none exist, it considers returning ones; only as a last resort does it interrupt active robots by queuing tasks behind their current job.
This triage model dramatically reduces computational overhead. In standard greedy schemes, every robot’s cost-to-task must be reevaluated at each assignment cycle—even if most are minutes away from being available. That’s wasteful. Here, only the relevant subset is examined. Simulations confirm the benefit: as fleet size grew from 8 to 16 units, task assignment time in conventional greedy methods climbed steadily. In the new approach? It remained flat—often under 15 milliseconds—even with larger fleets. For real-time systems operating under hard scheduling deadlines, this predictability is as valuable as raw speed.
But the real innovation lies in the third layer: anticipatory dispatch. Airports don’t generate random baggage surges. Checked luggage follows predictable statistical patterns—most arrivals cluster around flight departure windows, with inter-arrival times well-modeled by Poisson distributions. The researchers exploited this regularity. Their scheduler doesn’t wait passively for the next bag to show up. Instead, when the system predicts—based on historical λ (arrival rate) and μ (service rate) parameters—that a new task is imminent and idle robots are available, it pre-positions them near check-in zones. Similarly, robots flagged as “returning” (state 1) may be rerouted en route to intercept the next anticipated job, bypassing the staging area entirely. In effect, the system creates a “just-in-time” buffer of ready assets.
Think of it like a restaurant manager who, seeing the lunch rush approaching on a reservations app, instructs idle waitstaff to proactively stand by at the host stand—not back in the break room. Or like a subway system that deploys “shadow trains” during peak hours to absorb sudden passenger surges. This predictive readiness eliminates the typical lag between task detection and robot dispatch. In optimal conditions—when traffic is moderate and predictable—the fleet can achieve near-zero wait time: the robot is already rolling toward the counter as the bag is tagged.
The results speak for themselves. In comparative simulations against three established methods—SimpleGreedy (naïve greedy), GR (batched greedy with fixed time windows), and a beat-synchronized Contract Net Protocol (CNP)—the improved algorithm consistently delivered lower end-to-end system runtime. When 100 bags arrived over a 12.3-minute window (a realistic load for a mid-sized check-in island), the new method completed all deliveries in 778.2 seconds—versus 880.2 for CNP, a 11.6% improvement. With 14 robots handling 103 tasks in 5 minutes, runtime dropped to 436.8 seconds, outperforming CNP by 10.9%. Even under heavy congestion—92 bags arriving rapidly—the system held a stable 8.4% advantage.
Importantly, these gains aren’t fragile. The team tested robustness across variable fleet sizes (8–16 robots), task volumes (34–122 bags), and arrival windows (6.7–20.2 minutes). Performance didn’t collapse at scale. In fact, the relative advantage often increased as conditions became more realistic—not less. That suggests the approach isn’t just theoretically elegant; it’s operationally resilient.
For airport operators, the implications are tangible. A 10% reduction in baggage system runtime doesn’t just shave minutes off a single flight’s turnaround—it improves on-time performance across an entire hub’s schedule. Faster processing means lower risk of misrouted luggage (a $2.5 billion annual cost to airlines, per SITA). It enables more flexible terminal layouts, since AMRs don’t require embedded guidance infrastructure. And it future-proofs investments: as demand grows, adding more robots yields near-linear scalability, unlike legacy belt systems that require massive civil works to expand.
Yet technical success is only half the story. The true test for any logistics algorithm is integration into human workflows. Here, the design’s simplicity becomes a strength. Unlike black-box AI solutions that require data scientists to tune neural nets or interpret opaque decisions, this enhanced greedy strategy remains transparent and auditable. Dispatchers can understand why a particular robot was chosen—its state, its estimated path cost, its proximity to the next likely demand zone. That explainability builds trust, accelerates adoption, and eases compliance with aviation safety protocols that demand traceable decision logs.
Moreover, the computational frugality is deliberate. All logic runs on standard industrial PCs—not GPU clusters or cloud APIs. That matters in airside environments where network latency, cybersecurity, and hardware certification are non-negotiable. There’s no dependency on external services that could fail during a thunderstorm or spectrum congestion. It’s edge intelligence in the truest sense: smart, self-contained, and sovereign.
This work also subtly shifts the conversation around autonomy in aviation logistics. Much of the field’s attention has fixated on perception (better LiDAR, stereo vision) and localization (centimeter-accurate SLAM). Those are essential, but they solve only half the problem. As this research demonstrates, coordination intelligence—how robots negotiate, prioritize, and anticipate as a collective—may offer even larger efficiency dividends. You can have the world’s most accurate robot, but if it’s waiting idle while another crosses half the terminal for a nearby job, the system fails. Optimizing the orchestration layer is where the next wave of productivity gains will emerge.
Already, the team’s industry partner—Civil Aviation Chengdu Logistics Technology Co., Ltd.—is preparing for pilot deployment. Early feedback from airport engineering teams highlights two unexpected benefits. First, the state-based dispatch model naturally balances robot utilization. Idle units get priority, preventing wear-and-tear hotspots on overworked units. Second, the anticipatory dispatch reduces “deadheading”—empty travel between drop-offs and pickups—by up to 18% in initial field tests, directly cutting energy consumption and maintenance cycles.
Looking ahead, the researchers hint at several extensions. One involves dynamic reclassification: what if a robot’s state shifts mid-cycle due to a sudden obstacle or passenger intrusion? Another explores hybrid fleets—mixing high-capacity carts for bulk transport with nimble single-bag units for premium lanes. And longer-term, they envision integrating flight manifest data to predict baggage volume per counter hours in advance, enabling even more proactive robot staging.
Still, challenges remain. Real terminals feature dynamic obstacles—not just fixed walls, but moving trolleys, cleaning crews, and errant travelers. Future iterations may fuse real-time camera feeds or UWB tracking to update cost maps on the fly. Battery management is another frontier: how to sequence tasks to avoid mid-mission recharging interruptions? The paper touches on this only lightly, but it’s a top concern for 24/7 operations.
Nonetheless, the core contribution stands firm: a lean, physics-respecting, human-aligned scheduling logic that proves you don’t need deep learning to achieve deep efficiency. In an era where AI hype often overshadows practical engineering, this work is a refreshing reminder that elegance, insight, and domain knowledge still drive real-world progress.
As global air travel rebounds post-pandemic—with IATA projecting 7.3 billion passengers by 2035—the race to modernize ground infrastructure is accelerating. Airports in Singapore, Istanbul, and Doha are already trialing AMR-based baggage solutions. Yet many deployments stall at the pilot phase, tripped up by scaling bottlenecks or erratic robot behavior under pressure. This new algorithm offers a path past that impasse—not with more sensors or faster chips, but with smarter rules.
After all, the goal isn’t just automation. It’s reliable automation. It’s systems that don’t just move bags, but anticipate need, adapt to chaos, and deliver consistently—even when the departure board flashes red and the terminal hums with urgency.
In that sense, this isn’t merely an algorithmic upgrade. It’s a step toward airports that don’t just process passengers, but serve them—with grace, speed, and invisible competence.
—
Jin Xie¹, Yong Xiang²,³, Xiuqing Yang²,³, Xin-Zhi Zhou¹
¹ College of Electronics and Information Engineering, Sichuan University, Chengdu 610065, China
² The Second Research Institute of Civil Aviation Administration of China, Chengdu 610065, China
³ Civil Aviation Chengdu Logistics Technology Company Limited, Chengdu 610065, China
Journal of Sichuan University (Natural Science Edition), Vol. 58, No. 4, July 2021
DOI: 10.19907/j.0490-6756.2021.042003