RECORD
Manipulating Trajectory Prediction with Backdoors
Machine Learning · Submitted 21 Dec 2023
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim In other words, the model has good benign performance but is vulnerable to backdoors.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
Autonomous vehicles ought to predict the surrounding agents' trajectories to allow safe maneuvers in uncertain and complex traffic situations. As companies increasingly apply trajectory prediction in the real world, security becomes a relevant concern. In this paper, we focus on backdoors - a security threat acknowledged in other fields but so far overlooked for trajectory prediction. To this end, we describe and investigate four triggers that could affect trajectory prediction. We then show that these triggers (for example, a braking vehicle), when correlated with a desired output (for example, a curve) during training, cause the desired output of a state-of-the-art trajectory prediction model. In other words, the model has good benign performance but is vulnerable to backdoors. This is the case even if the trigger maneuver is performed by a non-casual agent behind the target vehicle. As a side-effect, our analysis reveals interesting limitations within trajectory prediction models. Finally, we evaluate a range of defenses against backdoors. While some, like simple offroad checks, do not enable detection for all triggers, clustering is a promising candidate to support manual inspection to find backdoors.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | Autonomous vehicles ought to predict the surrounding agents' trajectories to allow safe maneuvers in uncertain and complex traffic situations. | background 1.00 | 0.04 | 0.02 | asserted 0.69 |
| s1 | As companies increasingly apply trajectory prediction in the real world, security becomes a relevant concern. | background 0.99 | 0.04 | 0.02 | asserted 0.96 |
| s2 | In this paper, we focus on backdoors - a security threat acknowledged in other fields but so far overlooked for trajectory prediction. | gap 0.90 | 0.13 | 0.04 | asserted 0.99 |
| s3 | To this end, we describe and investigate four triggers that could affect trajectory prediction. | contribution 0.77 | 0.86 | 0.03 | hedged possibility 0.55 |
| s4 | We then show that these triggers (for example, a braking vehicle), when correlated with a desired output (for example, a curve) during training, cause the desired output of a state-of-the-art trajectory prediction model. | result 0.92 | 0.94 | 0.35 | asserted 1.00 |
| s5 | In other words, the model has good benign performance but is vulnerable to backdoors. | result 0.61 | 0.89 | 0.50 | asserted 1.00 |
| s6 | This is the case even if the trigger maneuver is performed by a non-casual agent behind the target vehicle. | result 0.96 | 0.84 | 0.07 | asserted 1.00 |
| s7 | As a side-effect, our analysis reveals interesting limitations within trajectory prediction models. | result 0.78 | 0.91 | 0.04 | asserted 0.96 |
| s8 | Finally, we evaluate a range of defenses against backdoors. | contribution 0.68 | 0.94 | 0.04 | asserted 0.94 |
| s9 | While some, like simple offroad checks, do not enable detection for all triggers, clustering is a promising candidate to support manual inspection to find backdoors. | result 0.96 | 0.91 | 0.23 | hedged possibility 0.84 |
What was examined, and what was not
- arXiv listing abstract v2
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
Autonomous vehicles ought to predict the surrounding agents' trajectories to allow safe maneuvers in uncertain and complex traffic situations. As companies increasingly apply trajectory prediction in the real world, security becomes a relevant concern. In this paper, we focus on backdoors - a security threat acknowledged in other fields but so far overlooked for trajectory prediction. To this end, we describe and investigate four triggers that could affect trajectory prediction. We then show that these triggers (for example, a braking vehicle), when correlated with a desired output (for example, a curve) during training, cause the desired output of a state-of-the-art trajectory prediction model. In other words, the model has good benign performance but is vulnerable to backdoors. This is the case even if the trigger maneuver is performed by a non-casual agent behind the target vehicle. As a side-effect, our analysis reveals interesting limitations within trajectory prediction models. Finally, we evaluate a range of defenses against backdoors. While some, like simple offroad checks, do not enable detection for all triggers, clustering is a promising candidate to support manual inspection to find backdoors.
Human review
How the literature uses it
What this paper does with the arXiv papers it cites
What it cites · 3 sentences without a confident label
Uses the method or tool of 1 paper
-
Analogous to computer vision, we use clustering to identify backdoors [ref] .
Curie: A method for protecting SVM Classifier from Poisoning Attack · Clustering · 0.92
Uses data or values from 1 paper
-
We train and evaluate our model using the publicly available nuScenes [ref] dataset.
nuScenes: A multimodal dataset for autonomous driving · Experimental Setting · 0.96
Cites as background 1 paper
-
One such attack is a backdoor, where an attacker inserts a strong correlation between a trigger (backdoor) pattern and an attacker-specified target behavior in the training data [ref] .
Dangerous Cloaking: Natural Trigger based Backdoor Attacks on Object Detectors in the Physical World · Introduction · 0.94
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.