E2MN: human-inspired end-to-end mapless navigation with oscillation suppression and short-term memory

Abstract

Robotic navigation in unknown environments is challenging due to the lack of high-definition maps. Building maps in real time requires significant computational resources. Nevertheless, sensor data can provide sufficient environmental context for robots’ navigation. This paper presents an interpretable and mapless navigation method using only two-dimensional (2D) light detection and ranging (LiDAR), mimicking human strategies to escape from dead ends. Unlike traditional planners, which depend on global paths or vision-based and learning-based methods, requiring heavy data and hardware, our approach is lightweight and robust, and it requires no prior map. It effectively suppresses oscillations and enables autonomous recovery from local minimum traps. Experiments across diverse environments and routes, including ablation studies and comparisons with existing frameworks, show that the proposed method achieves map-like performance without a map—reducing the average path length by 50.51% when compared to the classical mapless Bug2 algorithm and increasing it by only 17.57% when compared to map-based navigation.

1. Introduction

“Where can a wine shop be found to drown his sad hours? A cowherd boy points to a cot amid apricot flowers.” This classical Chinese poem reflects that, long before the advent of high-definition maps, humans could still reach destinations by following general directions. Even in unfamiliar environments, humans navigate effectively using short-term memory. Typically, humans tend to move in a straight line toward their goal. When encountering a dead end or realizing that the target is blocked by an obstacle, they briefly remember that the path is inaccessible and seek alternatives. One of the primary goals of robotic navigation is to replicate such human behavior, making this human-like navigation strategy suitable for robots as well.

Traditional autonomous robot navigation relies on coarse-grained global planning over prior maps to guide the robot toward its goal. The navigation process usually includes two stages: global planning and local planning. First, a global planner provides a global solution based on prior maps, and then a local planner continuously refines the path under its guidance (Zhong et al., 2023). However, in fully unknown scenarios, such as forest rescue or battlefield operations, where only the approximate direction and distance are known, robots must navigate directly using perception without global maps. Under such conditions, the navigation task faces the following key challenges: 1. Planning a short and smooth path toward the goal without a prior map and with limited hardware; 2. Preventing the system from falling into local optima during navigation; 3. Ensuring safe, collision-free navigation without global information.

In previous research, various advances have been made in mapless navigation, and several influential studies have been published. Many existing methods have achieved navigation in relatively simple environments, and some have enabled escape from local minima. However, these approaches still have notable limitations. Traditional local planners and reinforcement learning methods that directly output linear and angular velocities can navigate non-maze scenarios but tend to fall into local optima and struggle to escape from dead ends. Human-inspired navigation techniques often rely on visual input, which imposes high hardware requirements, whereas two-dimensional (2D) light detection and ranging (LiDAR) sensors already capture sufficient environmental information. Therefore, this work adopts LiDAR-based navigation. Existing strategies for escaping from local minima generally fall into two categories. One constructs topological maps from perception data, which suffer from step size constraints (long steps may overlook narrow passages, while short steps increase memory costs). The other predicts temporary goals directly, offering better spatial efficiency, but often lacks interpretability and produces unreasonable goal selection. This paper adopts the temporary goal-based approach.