RAEM: a new way to explore multi-story buildings with quadruped robots

How does RAEM solve the problem of robot stair exploration?

Most existing exploration systems for land robots rely on flat representations that cannot adequately represent overlapping structures or connections between floors. In the case of multi-story buildings, this approach leads to errors in route planning and loss of topological consistency, especially on stairs. RAEM solves this problem by introducing a hybrid space representation - it combines a local tomographic map with a categorized three-dimensional grid. This allows the robot to analyze the terrain in real time while maintaining a global topological graph.

An important element of the system is the strategy for aligning to the center of the stairs, which reduces sudden changes in direction (yaw) during ascent. This is especially important for quadrupeds, which must maintain balance and precision with each step. Additionally, RAEM introduces a two-way path search mechanism - when a local disconnection occurs in the topological graph, the system can recover the previous route and continue exploration without having to replan from scratch.

How does the hybrid space representation work in RAEM?

RAEM uses a two-layer approach to model the environment. At the local level, the robot creates a tomographic map and a categorized 3D grid, which allow for accurate assessment of traversability near the robot - especially in areas with limited visibility, such as stairs. This data is then used to analyze the terrain and evaluate connections between points.

At the global level, the system incrementally builds an elevation-aware topological graph, which enables efficient planning of exploration between floors. Thanks to this, the robot does not have to store a complete three-dimensional map, which reduces computational load and allows for faster replanning of routes in real time. This architecture enables both local accuracy and global scalability.

Tests confirmed the ability to continuously explore a five-story staircase.

Research conducted in both simulations and on actual terrain has shown that RAEM is able to perform autonomous exploration of multi-story structures without losing coherence. A key result was the ability to continuously explore a five-story staircase - the system did not stop working, even when local LiDAR data was fragmented or insufficient.

The results confirm that the hybrid space representation and path recovery mechanisms work effectively in real-world conditions. This represents a significant step forward towards the application of quadruped robots in real-world scenarios, such as search and rescue, building inspection after disasters, or patrolling hard-to-reach areas.

Can RAEM be used in practice?

Although the system was presented as a research framework, its results indicate real possibilities for application. In particular, the ability to explore stairs and maintain topological consistency may be crucial for robots used in rescue operations, post-disaster inspections, or patrolling industrial buildings.

At the same time, it should be emphasized that RAEM has not yet been implemented in actual operating systems or fully tested in autonomous form on site. All experiments were conducted under controlled conditions - both simulated and in a laboratory environment. This means that despite its promising potential, the system requires further work on stability, robustness to sensor noise, and integration with other robotic modules.

In practice, the application of RAEM may be crucial in situations where access to buildings is difficult or dangerous for humans - e.g., after a flood, earthquake, or technical failure. Thanks to its ability to continuously explore stairs and maintain topological consistency, the robot can transmit up-to-date data on the condition of rooms, locate hazards, or search for people in need without the need for operator intervention. Although the system has not yet been implemented in fully autonomous operational devices, its architecture is designed with scalability and integration with other robotic modules in mind, such as object recognition or real-time communication.

In the future, RAEM could become the foundation for rescue systems based on quadruped robots that are capable of independently exploring complex multi-story environments without losing orientation or requiring constant supervision. This represents a significant step forward towards full autonomy of robots in difficult conditions where traditional exploration methods are insufficient.