Autonomous navigation in unknown, complex indoor environments remains challenging due to limited sensing range and severe partial observability. Conventional methods rely on local maps without foresight, causing dead-ends and long detours, while local goal selection based on Euclidean distance or frontier coverage fails to balance efficiency with directionality. To address these challenges, we propose ForexNav, a foresight-enabled exploratory navigation framework. To handle structural ambiguity in unseen regions, we introduce Foresight Hypothesis Fusion (FHF), which maintains multiple WGAN-based map predictions and reweights them via temporal evidence accumulation. A Traversability-aware A* search then quantifies predictive traversability on the fused map, enabling a multi-objective planner to synthesize path feasibility, kinodynamic conformity, monotonic-progress consistency, and geometric distance for optimal intermediate goal selection and dynamically consistent trajectory generation. Experiments in four simulated indoor scenes of up to 3,300 m<inline-formula><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula> demonstrate navigation success while reducing total travel time by 25.0% and improving average velocity by 13.3% over the strongest baseline, with path ratio improvements of 22.2% on average in large-scale environments (<inline-formula><tex-math notation="LaTeX">$\geq$</tex-math></inline-formula>2,000 m<inline-formula><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula>). Real-world deployment on a quadruped robot supports practical feasibility, and extension to a fixed-altitude micro-UAV further suggests preliminary cross-platform transferability.
Hong-Yu Song, Yun-Fang Ren, Ji-Gui Miao et al.· IEEE Robotics and Automation...· 0 citations
Vision--language--action (VLA) models acquire broad generalization through large-scale pretraining, yet adapting them to a new task and robot embodiment still requires post-training on newly collected data. Unlike pretraining, post-training targets task- and embodiment-specific adaptation, making it particularly sensitive to data quality. In practice, collected robot datasets often contain heterogeneous errors, including execution mistakes, sensor drift, and timestamp misalignment, which can impair post-training and policy performance. Manual inspection is costly, while existing data-cleaning methods are typically tailored to particular corruption types. To address these challenges, we introduce \textsc{RoboDrop}, a data-curation framework that audits supervision using local gradient compatibility measured along the training trajectory as a proxy for its effect on post-training performance. During a one-epoch warm-up run, RoboDrop scores each candidate sample online by comparing its gradient with those of task-semantic and visually matched validation samples. The resulting sample scores are aggregated at the episode level, and a simple automatic post-processing rule converts them into filtering decisions. We evaluate RoboDrop on controlled observation--action corruptions, naturally suboptimal demonstrations in simulation, and real-robot datasets containing non-expert collection errors. Across these settings, RoboDrop more accurately distinguishes unreliable demonstrations than prior methods, while post-training on the curated data consistently yields stronger downstream policies, with average real-robot rollout success rising from $35.0\%$ to $67.5\%$. These results establish training-trajectory-aware, context-conditioned supervision auditing as an effective approach to robust VLA post-training.
Runze Xu, Yuanfan Xu, Cui-Jie Xu et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
