Remember Smarter: Visual History Compressor and Hyperbolic Experience Space for Robotic Memory

2026-08-15

AI Summary

This study addresses the challenges of limited context windows and inefficient experience reuse in Vision-Language-Action models for long-horizon robotic policies by proposing a plug-and-play memory module. The method employs bidirectional spatial Mamba to compress visual history and integrates a hyperbolic Variational Autoencoder for hierarchical storage with asynchronous retrieval, enabling compact memory access without context expansion. By pioneering complementary branches for visual compression and hyperbolic experience representation, the approach ensures zero-blocking inference. Experimental results demonstrate that this method increases the success rate on LIBERO-Plus from 53.6% to 70.6%. Furthermore, real-world robot evaluations confirm significant enhancements in both memory retention and experience utilization capabilities, validating the effectiveness of the proposed architecture for long-horizon manipulation tasks.

Abstract

Long-horizon robot policies require compact access to recent observations and reusable experience without expanding the vision-language-action (VLA) context. We introduce Remember Smarter (RS), a plug-and-play module with complementary visual-history and hyperbolic experience-memory branches. Its visual branch compresses multi-view patch histories using bidirectional spatial Mamba and causal temporal Mamba, then exposes the resulting memory to action-facing hidden states through residual cross-attention while leaving the VLM visual-token stream unchanged. Its experience branch stores successful final-layer VLM states in a Poincare VAE space, organizes them hierarchically, and asynchronously converts retrieved experience into geodesic prompt tokens without blocking action inference. When adapted to pi0, RS increases total success on LIBERO-Plus from 53.6% to 70.6% and achieves substantial performance gains in real-robot experiments designed to evaluate memory retention and experience utilization.

Problem

• Long-horizon robot policies
• Vision-Language-Action (VLA)
• Robotic memory
• Context expansion
• Experience reuse

Innovation

• Visual History Compressor
• Hyperbolic Experience Space
• Bidirectional Spatial Mamba
• Poincare VAE
• Asynchronous Geodesic Prompting