paper
Planning-aligned Token Compression for Long-Context Autonomous Driving
paperactiveprovisional
planning-aligned-token-compression-for-long-context-autonomous-driving-ae09466e·1 events·first seen 9d agoAliases: Planning-aligned Token Compression for Long-Context Autonomous Driving
Co-occurring entities
More like this (12)
Context-Driven Incremental Compression for Multi-Turn Dialogue GenerationContext-Driven Incremental Compression for Multi-Turn Dialogue GenerationEnd-to-End Context Compression at ScaleContext-Driven Incremental CompressionVisual Verification Enables Inference-time Steering and Autonomous Policy ImprovementBeyond Uniform Tokens: Adaptive Compression for Time Series Language ModelsCLP: Collocation-Length Prediction for Zero-Loss Adaptive Multi-Token Inferencevisual-token compressionintra-frame token sparsificationSparse AutoencodersRe-imagining ISO 26262 in the Age of Autonomous Vehicles: Enhancing Controllability through Transferability and PredictabilityTrace Only What You Need: Structure-Aware On-Demand Hypergraph Memory for Long-Document Question Answering
Recent events (1)
COMPACT-VA: Planning-aligned token compression for long-context autonomous driving
Researchers introduce COMPACT-VA, a working memory framework using conditional VQ-VAE to compress extended temporal context in vision-action autonomous driving models. Compression is conditioned on historical trajectory and a learned planning intent derived from future trajectories during training, enabling end-to-end optimization without backbone modifications. On high-signal dynamic scenarios, the method achieves 68.3% success rate (>6% improvement) with 3.3x speedup and 2.7x memory reduction over uncompressed processing.