RB5 Low-Cost Explorer: Implementing Autonomous Long-Term Exploration on Low-Cost Robotic Hardware

TL;DR

This work tackles autonomous long-term exploration in unknown GPS-denied environments using a low-cost robotic platform. It formalizes exploration of a bounded volume $\mathcal{Q}\subseteq\mathbb{R}^3$ with obstacle sets $\mathcal{Q}^{O_i}$ and uses a 2D path function $\phi:\mathbb{R}^2\to\mathbb{R}$ guided by a path-following vector field to enable low update frequencies. The RB5 platform integrates rocker-bogie suspension, RGB-D sensing, a Jetson Xavier NX, LoRa-based remote intervention, and a ROS2 software stack implementing a mixed frontier- and sampling-based exploration with a RTAB-Map SLAM backend. Field experiments across indoor structured corridors, unstructured spaces, and outdoor tunnels demonstrate obstacle avoidance and long-term autonomy, with remote intervention enabling continued operation; the system is released as open-source to enable replication and further development.

Abstract

This systems paper presents the implementation and design of RB5, a wheeled robot for autonomous long-term exploration with fewer and cheaper sensors. Requiring just an RGB-D camera and low-power computing hardware, the system consists of an experimental platform with rocker-bogie suspension. It operates in unknown and GPS-denied environments and on indoor and outdoor terrains. The exploration consists of a methodology that extends frontier- and sampling-based exploration with a path-following vector field and a state-of-the-art SLAM algorithm. The methodology allows the robot to explore its surroundings at lower update frequencies, enabling the use of lower-performing and lower-cost hardware while still retaining good autonomous performance. The approach further consists of a methodology to interact with a remotely located human operator based on an inexpensive long-range and low-power communication technology from the internet-of-things domain (i.e., LoRa) and a customized communication protocol. The results and the feasibility analysis show the possible applications and limitations of the approach.

Figures (5)

• Figure 1: RB5 low-cost wheeled robotic explorer. A robot needs to explore its surroundings with fewer and cheaper sensors -- the picture illustrates RB5, our experimental wheeled robotic platform that carries an RGB-D camera and low-power computing hardware to derive an exploratory coverage path.
• Figure 2: Detail of our autonomous exploration methodology. The approach consists of the robot sampling the environment and searching for obstacles and unexplored areas. The approach clusters the two groups into vertex sets and builds candidate path functions. From these, it selects the trajectory w.r.t. a given cost function and iterates the operation at each step. In between the iterations, it tracks the trajectory, saving computational and sensing resources.
• Figure 3: Autonomy for different classes of mobile robots. Autonomy is reported in hours between the time the battery is fully charged to discharged for our RB5 explorer and other approaches that report the value in the literature. Even though the metric is use case- and battery-dependent, the data show that the reported autonomy for wheeled robots is higher than the reported autonomy for legged, combination of legged and aerial, and aerial robots.
• Figure 4: Results for a structured environment. Experimental results are reported for a structured indoor environment, a university hallway composed of four connected corridors for a total length of approximately 80 meters. The view includes the point cloud in Fig. \ref{['fig:1-3']} and the detail of the algorithm for obstacle avoidance and detection at successive time steps in Fig. \ref{['fig:1-1']} and \ref{['fig:1-2']}. The points in the point cloud are filtered to report one point every 250. The colors of the spheres in Figs. \ref{['fig:1-1']}--\ref{['fig:1-2']} indicate the proximity of an obstacle (orange indicates close proximity) and arrows the path-following vector field in Eq. (\ref{['eq:pfvf']}). The robot's trajectory is in red and red dots indicate SLAM's registration points.
• Figure 5: Results for an unstructured environment. Experimental results are reported for an unstructured indoor environment and an underground tunnel for a total length of approximately 100 meters. The view of the point cloud in Fig. \ref{['fig:2-1']} is filtered to report one point every 500. The detail of the algorithm for successive time steps is shown in Figs. \ref{['fig:2-2']}--\ref{['fig:2-3']}, similar to Fig. \ref{['fig:1']}.

Theorems & Definitions (2)

• Definition 2.1: Path function
• Definition 2.2: Coverage