Improving Collision-Free Success Rate For Object Goal Visual Navigation Via Two-Stage Training With Collision Prediction

TL;DR

This work addresses collisions in object-goal visual navigation with RGB inputs by introducing a collision-free objective and a two-stage training framework. In Stage 1, a collision-prediction module learns to anticipate collisions during exploration without penalty, while Stage 2 adds a collision penalty and uses the fixed predictor to guide collision-averse navigation toward the target. Across multiple state-of-the-art models in AI2-THOR, the approach yields substantial gains in collision-free success rate ($CF ext{-}SR$) and collision-free path efficiency ($CF ext{-}SPL$), outperforming other collision-avoidance strategies. The results demonstrate that decoupling collision prediction from policy optimization mitigates over-conservatism and improves practical navigation performance, with promising implications for real-world robotics and sim-to-real transfer.

Abstract

The object goal visual navigation is the task of navigating to a specific target object using egocentric visual observations. Recent end-to-end navigation models based on deep reinforcement learning have achieved remarkable performance in finding and reaching target objects. However, the collision problem of these models during navigation remains unresolved, since the collision is typically neglected when evaluating the success. Although incorporating a negative reward for collision during training appears straightforward, it results in a more conservative policy, thereby limiting the agent's ability to reach targets. In addition, many of these models utilize only RGB observations, further increasing the difficulty of collision avoidance without depth information. To address these limitations, a new concept -- collision-free success is introduced to evaluate the ability of navigation models to find a collision-free path towards the target object. A two-stage training method with collision prediction is proposed to improve the collision-free success rate of the existing navigation models using RGB observations. In the first training stage, the collision prediction module supervises the agent's collision states during exploration to learn to predict the possible collision. In the second stage, leveraging the trained collision prediction, the agent learns to navigate to the target without collision. The experimental results in the AI2-THOR environment demonstrate that the proposed method greatly improves the collision-free success rate of different navigation models and outperforms other comparable collision-avoidance methods.

Figures (4)

• Figure 1: The illustration of the two-stage training method with collision prediction. In the first stage, the collision prediction module is trained by supervising the collision state of the navigation agent. In the second stage, the learned collision prediction is inputted into the agent to help it learn navigation under collision penalty.
• Figure 2: Three sampled navigation paths of L-sTDElstde model before & after using the proposed training method. The failure actions that cause collisions are marked using red circles with $\times N$ denoting the number of failure actions in the same place. Target objects are marked with blue bounding boxes.
• Figure 3: The learning curves of CF-SR using different collision avoidance methods in the training set.
• Figure 4: The CF-SR learning curves of ablation studies in the training set.