In the rapidly evolving landscape of modern logistics, the deployment of intelligent robots has become a cornerstone for enhancing efficiency and automation. As an integral part of this technological shift, I focus on the navigation of intelligent robots in warehouse environments, where numerous robots operate simultaneously with high path selectivity. The key to ensuring efficient logistics lies in robust navigation methods that integrate multiple data sources. In this paper, I propose a novel information fusion navigation method for logistics handling intelligent robots, aiming to address challenges such as poor obstacle avoidance, suboptimal control performance, and lengthy navigation paths. By leveraging advanced filtering algorithms and data fusion techniques, this method enhances the navigation capabilities of intelligent robots, ultimately improving their performance in real-world applications.
The core of my approach revolves around information fusion, which combines data from various sensors to create a coherent and accurate representation of the robot’s environment. This process is critical for intelligent robots, as it reduces data redundancy and computational complexity, leading to more reliable navigation. I introduce an augmented Kalman filtering algorithm, specifically the rank Kalman filter (RKF), to predict the motion state of the intelligent robot. This algorithm incorporates rank statistics to improve prediction efficiency, making it suitable for nonlinear systems commonly encountered in robotics. The discrete system model for the intelligent robot is given by:
$$ Y_k = A_k + l(Y_{k-1}) $$
$$ G_k = B_k + J(Y_k) $$
where \(Y_k\) represents the n-dimensional motion state vector, \(G_k\) is the m-dimensional observation vector, \(l(Y_{k-1})\) and \(J(Y_k)\) are nonlinear functions, and \(A_k\) and \(B_k\) denote noise vectors. To enhance prediction, I use rank sampling to generate a set of sampling points based on prior distributions, expressed as:
$$ Y_{i,k-1} = \hat{y} + e_j \alpha_{p_j} \sqrt{P_{k-1}^{(c)}} $$
where \(Y_{i,k-1}\) is the i-th sample at time k, \(P_{k-1}\) is the error covariance matrix, \(c\) indexes matrix columns, \(\alpha_{p_j}\) represents standard distribution probability quantiles, and \(e_j\) is a correction coefficient satisfying \(\sum_{j=1, j \neq 1+m}^{2\rho+1} e_j \alpha_{p_j} = 0\), with \(\rho\) as the total number of sampling layers. The RKF algorithm decomposes the error covariance matrix via singular value decomposition (SVD):
$$ Q_{k|k} = L_{k|k} M_{k|k} H^T_{k|k} $$
where \(Q_{k|k}\) is a symmetric matrix, and \(L_{k|k}\), \(M_{k|k}\), and \(H^T_{k|k}\) are decomposition matrices. The sampling point set for robot motion is derived as:
$$ X_{i,k|k} = \hat{x}_{k|k} + e_j \beta_{p_j} L_{u,k|k} s_u $$
and the nonlinear motion state sampling points are given by \(X^*_{i,k|k} = f(X_{i,k|k}, \delta_k)\). From this, the estimated motion state and predicted error covariance matrix are computed:
$$ \hat{x}_{k|k} = \frac{1}{\rho} \sum_{i=1}^{2\rho} X^*_{i,k|k} $$
$$ Q_{k|k} = \frac{1}{\alpha} \sum_{i=1}^{2\rho} e^*_i \left( X^*_{i,k|k} – \hat{x}_{k|k} \right) \left( X^*_{i,k|k} – \hat{x}_{k|k} \right)^T + p_k $$
where \(e^*_i\) is the correction coefficient for the covariance matrix. This prediction step ensures accurate state estimation for the intelligent robot, laying the groundwork for effective data fusion.
To fuse navigation data from multiple sensors, I employ a normalization and weighting scheme. Assume the intelligent robot uses \(m\) sensors, each providing \(l\) measurements. The mean and standard deviation for each sensor are calculated as:
$$ x_i = \frac{1}{l} \sum_{j=1}^{l} x_{ij} $$
$$ \sigma_i = \sqrt{\frac{1}{l-1} \sum_{j=1}^{l} (x_{ij} – x_i)^2} $$
The overall estimate and standard deviation across sensors are:
$$ x_0 = \frac{1}{m} \sum_{i=1}^{m} x_i $$
$$ \sigma_0 = \sqrt{\frac{1}{m-1} \sum_{i=1}^{m} (x_i – x_0)^2} $$
I use the Grubbs test to eliminate invalid data from the intelligent robot’s motion state measurements. The statistic for this test is:
$$ h_i = \frac{x_i – x_0}{\sigma_0} $$
If the maximum or minimum value satisfies \(g_m \geq g_0\), the corresponding data point \(x_m\) is discarded. Next, for normal-type fuzzy set sensor measurements \(B_i\) and \(B_0\), the degree of closeness is defined as:
$$ \gamma_{B_i}(t) = \exp\left[ -\left( \frac{t – x_i}{\sigma_i} \right)^2 \right] $$
$$ \gamma_{B_0}(t) = \exp\left[ -\left( \frac{t – x_0}{\sigma_0} \right)^2 \right] $$
This closeness measure reflects the weight for data fusion. After normalizing all sensor data, the weight for each data source is:
$$ \omega_i = \frac{S(B_i, B_0)}{\sum_{i=1}^{m} S(B_i, B_0)} $$
where \(S(B_i, B_0)\) represents the similarity between measurements. The fused data result for the intelligent robot navigation is then:
$$ x = \sum_{i=1}^{m} \omega_i x_i $$
This fusion process ensures that the intelligent robot operates with comprehensive and accurate environmental data, enhancing navigation precision. The following table summarizes the key parameters and formulas used in the information fusion method for the intelligent robot:
| Component | Formula/Description | Purpose |
|---|---|---|
| Motion State Prediction | $$ Y_k = A_k + l(Y_{k-1}) $$ | Models nonlinear discrete system for intelligent robot |
| Rank Sampling | $$ Y_{i,k-1} = \hat{y} + e_j \alpha_{p_j} \sqrt{P_{k-1}^{(c)}} $$ | Generates sampling points for state estimation |
| Error Covariance Decomposition | $$ Q_{k|k} = L_{k|k} M_{k|k} H^T_{k|k} $$ | Enhances prediction stability in RKF algorithm |
| Data Fusion Weights | $$ \omega_i = \frac{S(B_i, B_0)}{\sum_{i=1}^{m} S(B_i, B_0)} $$ | Assigns weights to sensor data for fusion |
| Fused Data Result | $$ x = \sum_{i=1}^{m} \omega_i x_i $$ | Combines sensor data for navigation |
With the fused data, I proceed to path planning for the logistics handling intelligent robot. I adopt the artificial potential field method, which creates a virtual force field to guide the intelligent robot. In this field, the target generates an attractive force, while obstacles generate repulsive forces. The total potential function is defined as:
$$ E(p) = E_{\text{att}}(p) + E_{\text{rep}}(p) $$
where \(E_{\text{att}}(p)\) is the attractive potential and \(E_{\text{rep}}(p)\) is the repulsive potential. These are given by:
$$ E_{\text{att}}(p) = \frac{1}{2} \zeta \rho_t^2 $$
$$ E_{\text{rep}}(p) = \begin{cases} \frac{1}{2} \epsilon \left( \frac{1}{\rho} – \frac{1}{\rho_0} \right)^2, & \text{if } \rho \leq \rho_0 \\ 0, & \text{if } \rho > \rho_0 \end{cases} $$
Here, \(\zeta\), \(\epsilon\), \(\rho_0\), and \(\rho\) are constants; \(\rho_t(p) = \| p – p_t \| = \sqrt{(x – x_t)^2 + (y – y_t)^2}\) is the distance between the intelligent robot and the target, and \(\rho(p) = \| p – p_0 \| = \sqrt{(x – x_0)^2 + (y – y_0)^2}\) is the distance between the intelligent robot and an obstacle. The virtual force field, as the negative gradient of the potential function, is:
$$ D(p) = -\nabla E(p) = D_{\text{att}}(p) + D_{\text{rep}}(p) $$
where \(D_{\text{att}}(p)\) is the attractive force and \(D_{\text{rep}}(p)\) is the repulsive force. By synthesizing these forces, the intelligent robot plans a global optimal path that avoids obstacles efficiently. This approach is particularly effective for intelligent robots operating in cluttered warehouse environments.
Next, I address the control of the intelligent robot’s movement through trajectory tracking. The kinematic model of the intelligent robot is expressed as:
$$ \begin{bmatrix} \dot{x} \\ \dot{y} \\ \dot{\lambda} \end{bmatrix} = \begin{bmatrix} \cos \gamma & 0 \\ \sin \gamma & 0 \\ 0 & 1 \end{bmatrix} \begin{bmatrix} v \\ w \end{bmatrix} $$
where \([v, w]\) is the composite control vector, with \(v\) as the linear velocity and \(w\) as the angular velocity of the intelligent robot. The standard motion trajectory is \(X_{\text{ref}} = f(t)\). To minimize tracking error, control effort, and control increments, I develop a predictive control model that includes a correction term for system disturbances:
$$ X(l+1|l) = X(l|l) + T_0 d(X(l|l), E(l|l)) + \vartheta E(l|l) $$
$$ \vdots $$
$$ X(l+N_p|l) = X(l+N_p-1|l) + T_0 d(X(l+N_p-1|l), E(l+N_p|l)) + \vartheta E(l+N_c|l) $$
where \(\vartheta\) is the error correction term, and \(E(l|l)\) is the control object. This model enables real-time control through rolling optimization and feedback correction, ensuring precise trajectory tracking for the intelligent robot. The control parameters are optimized to maintain the intelligent robot on the planned path, enhancing navigation accuracy.
To validate the proposed method, I conduct simulations comparing it with existing approaches, such as magnetic field-based navigation and visual navigation methods. The experiments focus on obstacle avoidance effectiveness, control performance for linear and angular velocities, and navigation path distance. For obstacle avoidance, I simulate a warehouse environment with multiple obstacles and evaluate how each method navigates the intelligent robot. The results show that my information fusion method allows the intelligent robot to maintain a safe distance from obstacles, often passing between them, whereas other methods lead to paths too close to obstacles or longer detours. This superior performance stems from the comprehensive data fusion, which reduces computational complexity and provides accurate environmental awareness for the intelligent robot.
In terms of control performance, I compare the linear velocity \(v\) and angular velocity \(w\) of the intelligent robot under different methods against ideal values. The proposed method demonstrates close alignment with ideal data, as summarized in the table below. This indicates effective control of the intelligent robot’s movements, crucial for smooth navigation. In contrast, other methods show significant deviations, leading to suboptimal navigation.
| Method | Linear Velocity Control (vs. Ideal) | Angular Velocity Control (vs. Ideal) | Remarks |
|---|---|---|---|
| Proposed Information Fusion | Close match, minimal error | Close match, minimal error | Optimal for intelligent robot navigation |
| Magnetic Field-Based Method | Large deviations observed | Moderate deviations | Poor control for intelligent robot |
| Visual Navigation Method | Significant errors | Significant errors | Inconsistent for intelligent robot |
For navigation path distance, I measure the total length of paths planned by each method in the same environment. The proposed method yields the shortest path, as it efficiently integrates data to optimize route planning for the intelligent robot. This reduces travel time and energy consumption, enhancing overall logistics efficiency. The comparison highlights the advantages of information fusion in enabling intelligent robots to navigate complex spaces effectively.
In conclusion, the information fusion navigation method for logistics handling intelligent robots proposed in this paper addresses key challenges in robot navigation. By incorporating rank Kalman filtering for state prediction and robust data fusion techniques, it enhances obstacle avoidance, control performance, and path optimization for intelligent robots. The simulations confirm its superiority over existing methods, demonstrating shorter navigation paths, better velocity control, and improved obstacle avoidance. This method contributes to the advancement of intelligent robot systems in logistics, paving the way for more autonomous and efficient operations. Future work could explore real-time implementation and scalability for large-scale intelligent robot networks in dynamic environments.
The integration of information fusion into intelligent robot navigation represents a significant step forward in robotics. As intelligent robots become more prevalent in logistics, methods like this will be essential for ensuring reliability and efficiency. By continuously refining these techniques, we can unlock new potentials for intelligent robots, making them smarter and more adaptable to complex tasks. The journey of intelligent robots in transforming industries is just beginning, and with innovative approaches, their impact will only grow.