💻 computer science

Enhanced Unscented Kalman Filter-Based SLAM in Dynamic Environments: Euclidean Approach

This paper proposes an innovative Euclidean-based Unscented Kalman Filter approach for SLAM in dynamic environments that effectively mitigates the disruptive impact of moving landmarks, outperforming conventional algorithms in both simulated benchmarks and realistic mapping tasks.

Original authors: Masoud Dorvash, Ali Eslamian, Mohammad Reza Ahmadzadeh

Published 2026-06-24

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Masoud Dorvash, Ali Eslamian, Mohammad Reza Ahmadzadeh

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 robot trying to navigate a busy room while drawing a map of its surroundings. This is called SLAM (Simultaneous Localization and Mapping). To do this, the robot looks for "landmarks"—fixed objects like a chair, a table, or a door corner—to figure out where it is and where it's going.

The problem arises when the room isn't empty. What if there are people walking around, or a dog running across the floor? To a robot, these moving things look like landmarks that are suddenly teleporting to new spots. If the robot tries to use these moving objects to build its map, it gets confused, the map becomes a mess, and the robot loses its way.

This paper presents a new, clever trick to help the robot ignore the moving chaos and focus only on the stable, stationary objects.

The Core Idea: The "Distance Check"

The authors propose a method using a mathematical tool called the Unscented Kalman Filter (UKF). Think of the UKF as a very smart guesser that predicts where the robot will be next based on its speed and direction.

Here is the simple logic the new method uses, explained with an analogy:

Imagine you are walking through a park. You spot a bench (a landmark).

Step 1: You measure the distance to the bench.
Step 2: You take a step forward. Based on your step size and direction, you predict where the bench should be relative to you if it hadn't moved.
The Check: You look at the bench again and measure the actual distance.
- Scenario A (Stationary): Your prediction matches the actual distance. The bench is still there, just where you expected. Verdict: "This is a real landmark. I'll keep it in my map."
- Scenario B (Moving): You predicted the bench should be 5 meters away, but when you look, it's actually 8 meters away. Verdict: "This object moved! It's not a reliable landmark. I'll throw it out of my map."

The paper calls this an Euclidean Approach, which is just a fancy way of saying they are using simple geometry (measuring straight-line distances) to spot the difference between what should happen and what did happen.

Why This Matters

The paper argues that traditional robots often get tripped up by moving objects. They try to map everything they see, including the moving people, which causes the map to warp and the robot to get lost.

By using this "Distance Check," the robot can filter out the moving "noise" (like people or pets) and only trust the "signal" (the walls, chairs, and doors). This makes the robot's internal map much cleaner and its navigation much more accurate.

What the Authors Did to Prove It

The researchers didn't just talk about this; they built a simulation in a computer program (MATLAB) to test it. They created a virtual world with:

Waypoints: A path for the robot to follow.
Landmarks: Some fixed (like walls) and some moving (like a rolling ball).
Noise: They added "static" or errors to the data to make it feel like a real, imperfect world.

They ran the simulation hundreds of times, changing the number of moving objects and the length of the path.

The Results:

Better Accuracy: The new method made far fewer mistakes than the old, standard method. The robot stayed on its path much better.
Robustness: Even when there were many moving objects or lots of "noise" in the data, the new method held up well.
Speed: The method was fast enough to run in real-time (taking about 0.04 to 0.05 milliseconds per step), which is crucial for a robot that needs to make decisions instantly.

The One Catch (The "Circle" Problem)

The authors are honest about a tiny limitation. If a moving object happens to move in a perfect circle around the robot at the exact same distance the robot expects, the robot might get fooled and think the object is stationary. However, the authors note that this is a very rare, theoretical scenario that almost never happens in the real world.

Summary

In short, this paper introduces a simple but effective "lie detector" for robot maps. By constantly checking if the distance to an object matches the robot's prediction, the robot can instantly spot and ignore moving things. This allows the robot to build a stable, accurate map even in a busy, dynamic environment, making it safer and more reliable for real-world tasks like vacuuming a house or delivering packages in a warehouse.

Technical Summary: Enhanced Unscented Kalman Filter-Based SLAM in Dynamic Environments

Problem Statement
Simultaneous Localization and Mapping (SLAM) is a foundational technology for robotics, yet conventional algorithms, particularly those based on the Unscented Kalman Filter (UKF), are primarily optimized for static environments. In dynamic settings, the presence of moving landmarks introduces significant inconsistencies. These moving features can corrupt the map, lead to inaccurate robot pose estimations, complicate data association, and increase computational load by forcing the system to process unstable information. While the UKF is known for its robustness in nonlinear estimation compared to the Extended Kalman Filter (EKF), standard UKF-SLAM implementations struggle to distinguish between stationary and moving landmarks, often treating dynamic objects as static, which degrades performance.

Methodology
The paper proposes a novel, low-complexity approach to handle dynamic environments by integrating a Euclidean distance-based detection mechanism into the UKF-SLAM framework. The core logic relies on the principle that a stationary landmark's distance from the robot should remain consistent with geometric predictions based on the robot's motion.

Distance Comparison: At each time step $k$ , the robot measures the distance to a landmark. At the subsequent step $k+1$ , the system estimates what the distance should be if the landmark were stationary, using the robot's previous position, the distance traveled, and the angle of travel (derived from the robot's compass).
Thresholding: The algorithm calculates the absolute difference between the estimated distance ( $\hat{d}_{r,i}^{k+1}$ $\hat{d}_{r, i}^{k + 1}$ ) and the actual measured distance ( $d_{r,i}^{k+1}$ $d_{r, i}^{k + 1}$ ).
- If $|\hat{d}_{r,i}^{k+1} - d_{r,i}^{k+1}| \leq \epsilon$ (where $\epsilon$ is a predefined threshold), the landmark is classified as stationary and included in the state vector for the UKF update.
- If the difference exceeds the threshold, the landmark is identified as moving and excluded from the state vector to prevent map corruption.
Limitations: The authors acknowledge a theoretical limitation: if a moving landmark travels in a circular path with a radius matching the estimated distance (within the error margin $\epsilon$ ), the algorithm may misclassify it as stationary. However, the authors note this scenario is rare in practical applications.

Key Contributions

Euclidean-Based Filtering: The introduction of a simple, geometric method to differentiate between stationary and moving landmarks without requiring complex prior knowledge of the environment or heavy computational overhead.
Enhanced UKF-SLAM: A modified UKF-SLAM algorithm that actively filters out moving landmarks, thereby improving the consistency of the map and the accuracy of robot localization.
New Dataset: The creation and introduction of a specific dataset designed for filter-based SLAM algorithms in dynamic environments, intended to serve as a benchmark for future research.
Comprehensive Evaluation: A rigorous evaluation framework comparing the proposed method against conventional UKF-SLAM across varying numbers of moving landmarks, stationary landmarks, and waypoints.

Results
The proposed algorithm was evaluated through MATLAB simulations and a realistic mapping scenario. Key findings include:

Accuracy: The proposed method consistently demonstrated lower localization errors (measured via Integral Absolute Error) compared to the conventional UKF-SLAM across all tested scenarios, including those with a high proportion of moving landmarks.
Robustness: The algorithm showed greater robustness to variations in the number of landmarks and different trajectory types. While performance improved with shorter step lengths (smoother trajectories) for both methods, the proposed method maintained a distinct advantage.
Computational Efficiency: Despite the added logic for moving landmark identification, the computational time remained low. The proposed algorithm averaged approximately 0.05 ms per step, compared to 0.04 ms for the conventional method, indicating that the reduction in state vector size (by excluding moving landmarks) offsets the cost of the detection logic.
Real-World Simulation: In a realistic simulation where robot observations were treated as dense landmarks, the proposed method successfully mapped the environment with a single moving object, outperforming competing methods.

Significance and Claims
The paper claims that this approach offers a "simple yet effective" solution to the challenge of dynamic environments in SLAM. By mitigating the disruptive impact of moving landmarks, the method enhances the reliability of filter-based SLAM systems. The authors position their work as a step toward redefining the field of filter-based SLAM in dynamic contexts, an area often overlooked in favor of visual-based or deep learning approaches. They emphasize that their method does not require prior environmental knowledge and is practical for real-world applications, particularly for robots with limited computational resources. The paper concludes that while further improvements are needed for real-world deployment, the proposed algorithm demonstrates the efficacy of filter-based approaches in dynamic scenarios and provides a solid foundation for future enhancements.