Robust Navigation with Language Pretraining and Stochastic Sampling

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 1494–1499, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 1494 Robust Navigation with Language Pretraining and Stochastic Sampling Xiujun Li♠♦ Chunyuan Li♦ Qiaolin Xia♣ Yonatan Bisk♠♦♥ Asli Celikyilmaz♦ Jianfeng Gao♦ Noah A. Smith♠♥ Yejin Choi♠♥ ♠Paul G. Allen School of Computer Science & Engineering, University of Washington ♣Peking University ♦Microsoft Research AI ♥Allen Institute for Artificial Intelligence {xiujun,ybisk,nasmith,yejin}@cs.washington.edu xql@pku.edu.cn {xiul,chunyl,jfgao}@microsoft.com Abstract Core to the vision-and-language navigation (VLN) challenge is building robust instruction representations and action decoding schemes, which can generalize well to previously un- seen instructions and environments. In this pa- per, we report two simple but highly effective methods to address these challenges and lead to a new state-of-the-art performance. First, we adapt large-scale pretrained language mod- els to learn text representations that generalize better to previously unseen instructions. Sec- ond, we propose a stochastic sampling scheme to reduce the considerable gap between the ex- pert actions in training and sampled actions in test, so that the agent can learn to correct its own mistakes during long sequential action decoding. Combining the two techniques, we achieve a new state of the art on the Room-to- Room benchmark with 6% absolute gain over the previous best result (47% → 53%) on the Success Rate weighted by Path Length metric. 1 Introduction The vision-and-language navigation (VLN) task, learning to navigate in visual environments based on natural language instructions, has attracted interest throughout the artificial intelligence re- search community (Hemachandra et al., 2015; An- derson et al., 2018; Chen et al., 2019; Savva et al., 2019). It fosters research on multimodal represen- tations and reinforcement learning, and serves as a test bed for many real-world applications such as in-home robots. In the recent Room-to-Room (R2R) VLN chal- lenge (Anderson et al., 2018), most state-of-the- art methods are developed based on an encoder- decoder framework (Cho et al., 2014; Sutskever et al., 2014), where a natural language instruc- tion is represented as a sequence of words, and a navigation trajectory as a sequence of actions, Exposure Bias Generalization Training Unseen Env. Seen Env. Evaluation Teacher.Forcing Student.Forcing Challenge 2 Challenge 1 Figure 1: Two challenges in VLN. enhanced with attention (Anderson et al., 2018; Wang et al., 2019; Fried et al., 2018; Ma et al., 2019a). Two important components are shared by all VLN agents: (i) an Instruction Encoder that employs a language model (LM) for instruction understanding; and (ii) an Action Decoder, where an appropriate sequence-level training scheme is required for sequential decision-making. Each component faces its own challenges (see Figure 1). The first challenge is generalizing grounded natural language instruction understanding from seen to unseen environments. Specifically, in the R2R task, only 69% of bigrams are shared be- tween training and evaluation.1 Existing work leverages pretrained GloVe embeddings (Penning- ton et al., 2014) to help generalize. In computer vision, it has been shown that large-scale models …

Content truncated to 500 words