Energy-Efficient SLAM via Joint Design of Sensing, Communication, and Exploration Speed

To support future spatial machine intelligence applications, lifelong simultaneous localization and mapping (SLAM) has drawn significant attentions. SLAM is usually realized based on various types of mobile robots performing simultaneous and continuous sensing and communication. This paper focuses on analyzing the energy efficiency of robot operation for lifelong SLAM by jointly considering sensing, communication and mechanical factors. The system model is built based on a robot equipped with a 2D light detection and ranging (LiDAR) and an odometry. The cloud point raw data as well as the odometry data are wirelessly transmitted to data center where real-time map reconstruction is realized based on an unsupervised deep learning based method. The sensing duration, transmit power, transmit duration and exploration speed are jointly optimized to minimize the energy consumption. Simulations and experiments demonstrate the performance of our proposed method.

💡 Research Summary

The paper addresses the critical issue of energy consumption in lifelong simultaneous localization and mapping (SLAM) performed by mobile robots equipped with a 2‑D LiDAR and odometry. While most prior work treats sensing, motion, and wireless communication as separate subsystems, this study builds a unified system model that captures the intricate coupling among these three components. The robot continuously scans its environment with a 360° LiDAR (1° angular resolution) while moving along the perimeter of a square area. Raw point‑cloud data together with odometry measurements are cached locally and then transmitted wirelessly to an edge server where a DeepMapping‑based unsupervised deep neural network reconstructs an occupancy map in real time.

The authors formulate the total energy consumption as the sum of three terms: (i) communication energy (E_{\text{comm}} = \sum_{k=2}^{N_m+1} p_{tx,k} t_{\text{comm}}), (ii) LiDAR sensing energy (E_{\text{LiDAR}} = N_m E_L) (constant per scan), and (iii) mechanical motion energy (E_{\text{mech}} = p_e \cdot \frac{4(L-2e)}{v}), where the mechanical power model (p_e = \frac{1}{2}\kappa_1 v^3 + \kappa_2 v) captures both aerodynamic drag and rolling friction. The wireless link is modeled with a Rice‑distributed channel gain, free‑space path loss, and Shannon capacity, yielding a time‑varying transmission rate (R_k(t) = B \log_2!\bigl(1 + \frac{p_{rx,k}(t) |h_k|^2}{\sigma_k^2}\bigr)). The data payload per period is ((360 a_1 + 6 a_2)) bits, where (a_1) and (a_2) are the quantization bits for LiDAR and odometry respectively.

The core optimization problem (P1) seeks to minimize total energy with respect to transmit power (p_{tx,k}), communication‑time fraction (\rho), LiDAR sensing duration (t_{\text{sens}}), and robot speed (v), subject to (a) sufficient data transmission per period, (b) a hard deadline (T_{\max}), (c) positivity of power, (d) (0<\rho\le 1), and (e) a minimum number of LiDAR cycles (N_D). Lemma 1 proves that the optimal (\rho) is 1, i.e., allocating the entire sensing period to communication yields the lowest energy. Consequently, the optimal transmit power for each period can be expressed analytically (Eq. 27) as a function of distance, channel gain, and the required data volume.

With (\rho) fixed, the problem reduces to (P2), which only involves (t_{\text{sens}}) and (v). By approximating the number of periods (N_m \approx \frac{4(L-2e)}{v t_{\text{sens}}}) for large (N_D), the authors show that increasing (t_{\text{sens}}) monotonically reduces both motion and communication energy, while still satisfying the LiDAR hardware limits. The optimal sensing period is therefore (t_{\text{sens}}^\star = \frac{4(L-2e)v}{N_D}), i.e., the longest feasible scan that meets the minimum cycle requirement.

The speed (v) remains coupled to the communication power through the distance‑dependent path loss. Because a closed‑form solution for (v) is intractable, the authors derive an upper bound on total energy (E_{\text{total}}^{\text{up}}(v)) (Eq. 35) and analyze its derivative. The derivative is always positive for (v>0), indicating that total energy grows with speed. Hence, the optimal speed is the smallest value that still satisfies the deadline constraint (N_m t_{\text{sens}} \le T_{\max}).

Simulation results (Fig. 4) confirm that total energy decreases as the sensing period grows within realistic LiDAR limits, and that the proposed joint design outperforms baseline schemes that optimize each subsystem separately. Experiments on a real robot platform validate the analytical findings: the joint optimization reduces overall energy consumption by 10–30 % while maintaining comparable map reconstruction accuracy, as measured by the DeepMapping loss. The authors also construct a dataset from a square test area, partitioned into edge and corner subsets, demonstrating the method’s scalability across different map structures.

In summary, the paper makes three major contributions: (1) a comprehensive energy model that integrates sensing, communication, and mechanical motion for lifelong SLAM; (2) a tractable joint optimization framework that yields closed‑form solutions for communication time allocation and transmit power, and provides clear guidelines for sensing period and robot speed; (3) experimental validation on both simulation and physical robot platforms, showing substantial energy savings without sacrificing mapping performance. The work is particularly relevant for battery‑powered autonomous agents, long‑duration exploration robots, and industrial mobile platforms where energy efficiency is a primary design constraint.