NextBestPath: Efficient 3D Mapping of Unseen Environments

TL;DR

Active 3D mapping in unseen indoor environments is challenging due to short-sighted view planning and limited benchmarks. The authors propose Next-Best-Path (NBP), a unified model that predicts long-term goals via a value map $M_{c_t}$ and obstacle map $O_{c_t}$, guided by a Mapping Progress Encoder to plan efficient trajectories with Dijkstra-based path planning. They introduce AiMDoom, a Doom-based indoor dataset with four difficulty levels to stress-test mapping strategies, and demonstrate that NBP achieves state-of-the-art results on MP3D and AiMDoom, surpassing prior NBV methods by effectively balancing exploration and long-horizon planning. The work shows that joint prediction of coverage gains and obstacles, along with curriculum learning and online data collection, yields robust performance across diverse indoor geometries, though very hard scenes remain a challenge and warrant future global-optimization approaches.

Abstract

This work addresses the problem of active 3D mapping, where an agent must find an efficient trajectory to exhaustively reconstruct a new scene. Previous approaches mainly predict the next best view near the agent's location, which is prone to getting stuck in local areas. Additionally, existing indoor datasets are insufficient due to limited geometric complexity and inaccurate ground truth meshes. To overcome these limitations, we introduce a novel dataset AiMDoom with a map generator for the Doom video game, enabling to better benchmark active 3D mapping in diverse indoor environments. Moreover, we propose a new method we call next-best-path (NBP), which predicts long-term goals rather than focusing solely on short-sighted views. The model jointly predicts accumulated surface coverage gains for long-term goals and obstacle maps, allowing it to efficiently plan optimal paths with a unified model. By leveraging online data collection, data augmentation and curriculum learning, NBP significantly outperforms state-of-the-art methods on both the existing MP3D dataset and our AiMDoom dataset, achieving more efficient mapping in indoor environments of varying complexity.

Figures (9)

• Figure 1: Reconstruction results and trajectories of MACARONS guedon2023macarons and our NBP model. guedon2023macarons fails to fully map the environment in simple scenes (a), while our NBP model manages to capture the full scene (b), even in much more complex geometry (c).
• Figure 2: Maps from our AiMDoom dataset. The AiMDoom dataset includes four levels of geometric complexity with various textures.
• Figure 3: Overview of the proposed next-best-path (NBP) framework. The model (left, see Section \ref{['sec:method_model']}) predicts a value map of coverage gain and an obstacle map, which are used for decision making (right, see Section \ref{['sec:method_decisionmaking']}) to obtain a next-best path.
• Figure 4: Comparison of our NBP method with the state-of-the-art MACARONS method. Both methods start from the same initial pose, marked in deep blue. We also include a demonstration video of active mapping using our method in the supplementary materials.
• Figure 5: Comparisons of different spatial ranges for value map prediction.