🤖 AI

Industrial Internet Robot Collaboration System and Edge Computing Optimization

This paper presents an industrial robot collaboration system that integrates a three-layer neural network with simulated annealing for global path planning and a fuzzy controller for local execution, optimized via edge computing to achieve robust, collision-free navigation with high precision and low latency.

Original authors: Haopeng Zhao, Dajun Tao, Tian Qi, Jingyuan Xu, Zijie Zhou, Lipeng Liu

Published 2026-01-30

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Haopeng Zhao, Dajun Tao, Tian Qi, Jingyuan Xu, Zijie Zhou, Lipeng Liu

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 busy warehouse where a fleet of delivery robots needs to zip around shelves, avoid forklifts, and drop off packages without bumping into anything. This paper is about teaching these robots how to navigate that chaotic environment safely and efficiently.

Here is the story of how the authors solved this problem, broken down into simple concepts:

1. The Problem: A Robot with a "Mind of Its Own"

In the real world, robots aren't perfect. If you tell a robot to drive straight, it might drift slightly because the floor is slippery or the wheels aren't perfectly round. In a crowded warehouse, if a robot drifts even a little, it could crash into a shelf or another robot.

The authors realized that to fix this, you can't just tell the robot "go there." You have to understand how the robot moves (its physics) and then give it a plan that accounts for its imperfections.

2. The Strategy: Breaking the Journey into Three "Modes"

Instead of trying to solve the whole trip at once, the authors broke the robot's movement down into three simple behaviors, like changing gears in a car:

Obstacle Avoidance: "I see a wall! Turn away!" (This uses the angle of the obstacle to decide how sharply to turn).
Target Approaching: "I'm far from the goal; let's speed up, but not too fast." (This adjusts speed based on how far away the destination is).
Target Turning: "I'm close to the goal; let's slow down and turn precisely to face it."

By switching between these three "modes," the robot can handle complex situations more naturally.

3. The Map Maker: The "Energy" Planner

To figure out the best route from Point A to Point B, the team used a digital brain (a neural network) combined with a clever search method called Simulated Annealing.

Think of this like a hiker trying to find the lowest point in a foggy mountain valley:

The Goal: Find the path with the "lowest energy." In this context, "energy" means a mix of two things: short distance (walking less) and safety (staying far from obstacles).
The Trick: Sometimes, the hiker might get stuck in a small dip (a local trap) thinking it's the bottom. Simulated Annealing is like giving the hiker a little "shake" or "jump" now and then. This allows the robot to jump out of small traps and find the true best path (the global minimum) rather than just settling for a "good enough" one.

4. The Steering Wheel: The "Fuzzy" Controller

Once the robot has a map, it still needs to drive. This is where the Fuzzy Controller comes in.

Imagine you are driving a car and you start drifting toward the right lane. You don't need a math degree to know you should turn the wheel slightly left. You just "feel" the error.

The robot does the same thing. It constantly checks two things: Am I facing the right way? (Angle error) and Am I in the middle of the lane? (Lateral error).
If the robot is off by a few degrees or a few inches, the fuzzy controller acts like an experienced driver's hands, making tiny, smooth adjustments to the speed of the left and right wheels to nudge the robot back on track.

5. The "Edge" Advantage: Thinking Locally

The paper also mentions Edge Computing. Imagine if the robot had to send a photo of every obstacle to a giant server in a different city, wait for the server to calculate the answer, and then send it back. That would take too long, and the robot might crash in the meantime.

Instead, this system does the heavy thinking on the robot itself (or on a computer right next to it). It's like having a co-pilot sitting in the car with you rather than calling a navigation expert on the phone. This makes the robot react instantly to changes.

6. The Results: The Test Drive

The authors tested this system in a computer simulation (using software called Matlab 2024) and compared it against two other common methods.

Accuracy: The new method kept the robot within 5 centimeters (about 2 inches) of the perfect path.
Speed: It corrected its course in just 10 milliseconds (faster than a human blink).
Efficiency: The paths it chose were shorter than the other methods. For example, after 500 tries, the new method found a path of 106 meters, while the others took 117–118 meters.

Summary

In short, this paper presents a robot navigation system that:

Understands how the robot physically moves.
Plans the best route by balancing "shortest path" vs. "safest path" using a smart search algorithm.
Steers the robot in real-time using a "fuzzy" logic system that acts like an experienced driver correcting small mistakes.
Thinks locally to avoid delays.

The result is a robot that can navigate a messy, changing industrial environment with high precision and speed, avoiding collisions better than previous methods.

Technical Summary: Industrial Internet Robot Collaboration System and Edge Computing Optimization

Problem Statement
In industrial Internet environments, such as warehousing and flexible manufacturing, mobile robots face significant challenges in generating collision-free global routes. These challenges arise from stochastic obstacle layouts and random perturbations in commanded linear and angular velocities. While existing approaches optimize for minimum distance or energy, they often struggle with local minima and execution disturbances in complex, cluttered environments. Furthermore, high-precision navigation requires strict adherence to nonholonomic motion constraints while maintaining stable tracking and collision avoidance. The paper also addresses the latency and traffic issues inherent in robot-server communication, proposing edge computing as a means to improve responsiveness.

Methodology
The proposed approach integrates dynamic modeling, energy-based global path planning, and fuzzy control for execution, supported by edge-side computation.

Dynamic Modeling and Behavior Decomposition:
The authors model a differential-drive robot with nonholonomic constraints. The dynamic equation is initially established based on pose ( $x, y, \theta$ ) and control inputs, then simplified into a kinematic form where linear velocity ( $\eta$ ) and angular velocity ( $v$ ) drive the motion. To enhance controllability, robot motion is decomposed into three parameterized behaviors:
- Obstacle Avoidance: Adjusts angular velocity based on deviation angles and obstacle bearing, while modulating linear velocity based on minimum obstacle distance.
- Target Approaching: Limits linear velocity based on the distance to the goal and maximum speed, while retaining obstacle-related terms.
- Target Turning: Sets angular velocity based on the difference between goal and obstacle bearings, bounding linear velocity.
Global Path Planning (Energy Function & Simulated Annealing):
Global path planning is formulated as a constrained optimization problem. The model is converted into a weighted energy function ( $F$ ) comprising a path-length term ( $F_l$ ) and a collision-penalty term ( $F_z$ ). A three-layer neural network represents the planning model. To solve for the optimal path and mitigate local minima, the authors employ Simulated Annealing. This algorithm iteratively generates neighboring paths, evaluates energy changes, and accepts inferior solutions with a probability that decreases as the "temperature" lowers, converging toward a low-energy global path.
Global Path Control (Fuzzy Logic):
During execution, a fuzzy controller corrects deviations. The inputs are heading angle deviation (set threshold: $\pm10^\circ$ ) and lateral offset error (set threshold: $\pm100$ mm). The output is the differential speed between the left and right wheels. The controller utilizes triangular membership functions for error correction and defuzzification based on expert rules to output crisp control actions.
Edge Computing Integration:
The paper discusses moving computation to the edge to reduce robot-server traffic and latency, facilitating timely sharing of environmental information among collaborative robots.

Key Contributions

Behavior-Based Parameterization: A novel decomposition of robot motion into obstacle avoidance, target turning, and target approaching behaviors to parameterize control variables under nonholonomic constraints.
Hybrid Planning Architecture: The combination of a neural network-based planning model with Simulated Annealing to search for near-global minima in complex energy landscapes.
Error-Correcting Fuzzy Controller: A specific fuzzy control strategy that uses heading and lateral-offset errors to rapidly adjust wheel-speed differentials, ensuring the robot stays within defined deviation thresholds.
Edge-Optimized Execution: A framework that integrates edge computing to address latency and bandwidth constraints in industrial collaboration.

Experimental Results
The method was evaluated using Matlab 2024 simulations, comparing the proposed approach against two baselines: a visual-servo generalized-constraint planner and a dynamic/static safety-field planner.

Path Deviation: The proposed method maintained deviations within $\pm5$ cm and achieved convergence within 10 ms, outperforming the baselines in both stability and speed of convergence.
Path Length: As iterations increased (up to 500), the proposed method consistently generated shorter paths. At 500 iterations, the proposed method achieved a path length of 106 m, compared to 118 m (visual-servo) and 117 m (safety-field). The mean path length for the proposed method was 114 m, significantly shorter than the baselines (132 m and 130 m, respectively).
Robustness: The simulations demonstrated improved robustness in global navigation under stochastic conditions.

Significance and Limitations
The paper claims that this approach improves the robustness of global navigation in practical industrial settings by effectively balancing path optimality with collision avoidance and tracking stability. The integration of edge computing is highlighted as a critical factor for reducing latency in collaborative environments.

The authors modestly acknowledge limitations, noting the system's dependence on specific modeling assumptions, the use of hand-crafted fuzzy membership functions and rules, and sensitivity to sensor noise or abrupt environmental changes. Future work is identified as incorporating uncertainty-aware sensing, adaptive tuning of fuzzy parameters, and validation through real-robot experiments.