Research map / Label-efficient and robust learning / Recognition and long-tail
BeyondSight: Object Permanence for End-to-End Autonomous Driving
Sandro Papais, Letian Wang, Mudit jain, Behnaz Rezaei, Steven Waslander
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add permanence-aware temporal prior, fusion and unobservable supervision to SparseDrive | SparseDrive · nuScenes-Permanence val | mAP_unobs: 0 -> 0.249 | Improved |
| add BeyondSight to SparseDrive | SparseDrive · nuScenes val planning | L2 avg: 0.61 -> 0.54 | Improved |
| add BeyondSight to SparseDrive | SparseDrive · nuScenes val perception | mAP: 0.415 -> 0.427 | Improved |
| add temporal prior decoder without permanence annotations (ablation) | SparseDrive · standard nuScenes annotations | mAP_unobs: 0.000 -> 0.000 | No clear effect |
| add posterior fusion without permanence annotations (ablation) | SparseDrive + temporal prior · standard nuScenes annotations | mAP_obs: 0.417 -> 0.413 | Hurt |
| add temporal prior with permanence annotations (ablation) | SparseDrive · nuScenes-Permanence | mAP_unobs: 0.021 -> 0.126 | Improved |
| add posterior fusion decoder (ablation) | temporal prior model · nuScenes-Permanence | mAP_unobs: 0.126 -> 0.194 | Improved |
| add unobservable-actor supervision (ablation) | prior + fusion · nuScenes-Permanence | mAP_unobs: 0.194 -> 0.213 | Improved |
| add observability loss to full model (ablation) | BeyondSight · nuScenes-Permanence | mAP_unobs: 0.213 -> 0.249 | Improved |
| use BeyondSight on high-occlusion subset | SparseDrive · High-occlusion subset (600 samples) | L2 avg / CR avg: 0.64 -> 0.56 L2; 0.163 -> 0.08 CR | Improved |
| BeyondSight on observable actors' forecasting | SparseDrive · nuScenes-Permanence observable | minADE_obs: 0.610 -> 0.625 | Hurt |
| longer occlusion duration (ablation) | BeyondSight · nuScenes-Permanence | mAP_unobs: 0.307 (0-2s) -> 0.219 (4-6s) | Hurt |
Closest papers by meaning
- Future-Aware End-to-End Driving: Bidirectional Modeling of Trajectory Planning and Scene Evolution (NeurIPS 2025 · similarity 0.89)
- Driving in the Occupancy World: Vision-Centric 4D Occupancy Forecasting and Planning via World Models for Autonomous Driving (AAAI 2025 · similarity 0.88)
- Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.87)
- Drive-R1: Bridging Reasoning and Planning in VLMs for Autonomous Driving with Reinforcement Learning (AAAI 2026 · similarity 0.87)
- Don't Shake the Wheel: Momentum-Aware Planning in End-to-End Autonomous Driving (CVPR 2025 · similarity 0.87)
- DriveAgent-R1: Advancing VLM-based Autonomous Driving with Active Perception and Hybrid Thinking (ICLR 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
