Research map / Adversarial robustness and security / Certified and empirical robustness
AdvDO: Realistic Adversarial Attacks for Trajectory Prediction
Yulong Cao, Chaowei Xiao, Anima Anandkumar, Danfei Xu, Marco Pavone
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 |
|---|---|---|---|
| attack with Opt-init instead of corrected search* | AgentFormer w/ map · nuScenes val | ADE: 1.88 -> 3.39 | Improved |
| free the current dynamics (Opt-init) rather than fixing them (Opt-end) (ablation) | AgentFormer w/ map · nuScenes val | ADE: 2.23 -> 3.39 | Improved |
| Opt-init vs Opt-end on Trajectron++ w/o map (ablation) | Trajectron++ w/o map · nuScenes val | MR: 49.5% -> 76.8% | Improved |
| use unbounded search attack instead of curvature-bounded search* (ablation) | AgentFormer w/ map · nuScenes val | violation rate: 0% -> 10% | Hurt |
| remove the map input from AgentFormer (ablation) | AgentFormer · nuScenes val under Opt-init | ADE: 3.39 -> 3.81 | Hurt |
| evaluate with planning-aware weighting | AgentFormer w/ map · nuScenes val | PI-ADE: 1.38 -> 2.05 | Improved |
| adversarially train on Opt-init attack instead of search attack (ablation) | AgentFormer w/ map · nuScenes val | ADE on Opt-init attack: 2.81 -> 2.54 | Improved |
| adversarial training of any kind on clean data (ablation) | AgentFormer w/ map · nuScenes val benign | ADE: 1.83 -> 2.38 | Hurt |
| add AdvDO-generated data augmentation (ablation) | AgentFormer · nuScenes val benign | ADE: 1.83 -> 1.69 | Improved |
| compare generated trajectory sensitivity change of Opt-end versus search | AgentFormer w/ map · nuScenes val | Delta sensitivity: 2.33 -> 1.12 | Improved |
| attack of closed-loop planning with MPC rather than rule-based planner | Planner · nuScenes subset | collisions: 12/150 -> 7/150 | Improved |
| closed-loop instead of open-loop planning | MPC planner · nuScenes subset | off road: 43/150 -> 23/150 | Improved |
Closest papers by meaning
- On Adversarial Robustness of Trajectory Prediction for Autonomous Vehicles (CVPR 2022 · similarity 0.96)
- Enduring, Efficient and Robust Trajectory Prediction Attack in Autonomous Driving via Optimization-Driven Multi-Frame Perturbation Framework (CVPR 2025 · similarity 0.92)
- Adversarial Backdoor Attack by Naturalistic Data Poisoning on Trajectory Prediction in Autonomous Driving (CVPR 2024 · similarity 0.91)
- Certified Human Trajectory Prediction (CVPR 2025 · similarity 0.87)
- World Models as Adversaries: Multi-Agent Self-Play Fine-Tuning for Robust Motion Planning (NeurIPS 2026 · similarity 0.85)
- SlowTrack: Increasing the Latency of Camera-Based Perception in Autonomous Driving Using Adversarial Examples (AAAI 2024 · similarity 0.85)
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.
