M2I: From Factored Marginal Trajectory Prediction to Interactive Prediction

Sun, Qiao; Huang, Xin; Gu, Junru; Williams, Brian; Zhao, Hang

doi:10.1109/cvpr52688.2022.00643

Cited by 50 publications

(21 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Following the Waymo motion prediction challenge design, we train and evaluate our model using only the vehicle agents labeled as objects of interest. Table III shows the performance of our approach along with two new baselines: MotionCNN [10] and M2I [24]. We run the pre-trained models from the two baselines and keep the first 3s prediction for evaluation.…”

Section: B Quantitative Resultsmentioning

confidence: 99%

“…The performance of our model trained for 8s prediction on the interactive validation set is: minADE 6 = 1.59 and minFDE 6 = 3.67. As a comparison, the Waymo LSTM baseline achieves minFDE 6 = 6.07 for 8s prediction [30], and M2I [24] reports minFDE 6 = 5.49.…”

Section: B Quantitative Resultsmentioning

confidence: 99%

“…WIMP [20], CXX [22], TNT [23], GoalNet [15] and SGNet [17] first predict goal positions and then generate trajectories between the agent's last observed position and the predicted goals. IMAP [5] and M2I [24], instead of predicting trajectories only for target agents, perform prediction sequentially for all the agents. P2T [13] and PGP [25] take a reinforcement learning approach, with a reward-based model upon which a policy is trained to perform prediction.…”

Section: Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Context-Aware Timewise VAEs for Real-Time Vehicle Trajectory Prediction

Xu¹,

Hayet²,

Karamouzas³

2023

Preprint

View full text Add to dashboard Cite

Real-time, accurate prediction of human steering behaviors has wide applications, from developing intelligent traffic systems to deploying autonomous driving systems in both real and simulated worlds. In this paper, we present ContextVAE, a context-aware approach for multi-modal vehicle trajectory prediction. Built upon the backbone architecture of a timewise variational autoencoder, ContextVAE employs a dual attention mechanism for observation encoding that accounts for the environmental context information and the dynamic agents' states in a unified way. By utilizing features extracted from semantic maps during agent state encoding, our approach takes into account both the social features exhibited by agents on the scene and the physical environment constraints to generate map-compliant and socially-aware trajectories. We perform extensive testing on the nuScenes prediction challenge [1], Lyft Level 5 dataset [2] and Waymo Open Motion Dataset [3] to show the effectiveness of our approach and its state-of-the-art performance. In all tested datasets, ContextVAE models are fast to train and provide highquality multi-modal predictions in real-time.

show abstract

Section: B Quantitative Resultsmentioning

confidence: 99%

Section: B Quantitative Resultsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Context-Aware Timewise VAEs for Real-Time Vehicle Trajectory Prediction

Xu¹,

Hayet²,

Karamouzas³

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The RMSE measures the average displacement error between the predicted trajectories and their corresponding ground truth. This can be either the average error over the entire prediction horizon, reported as mADE, mean average displacement error in literature [11]- [14] or average of momentary error at different time points in the prediction horizon, reported as mFDE, mean final displacement error in literature [11]- [14]. A model with the lowest error is generally considered the better alternative.…”

Section: Introductionmentioning

confidence: 99%

Beyond RMSE: Do Machine-Learned Models of Road User Interaction Produce Human-Like Behavior?

Srinivasan

Lin

Antonello³

et al. 2023

IEEE Trans. Intell. Transport. Syst.

View full text Add to dashboard Cite

show abstract

“…However, these methods output the long-term control variables directly, which can be risky when there are HVs on the road. Learning-based approaches [2,3,16,21] obtain vehicles' driving strategies from open datasets and avoid collisions by predicting other cars' trajectories. The trajectories generated by such methods fail to express driving intentions explicitly and are therefore poorly understood by passengers or other road users.…”

Section: Introductionmentioning

confidence: 99%

Bringing Diversity to Autonomous Vehicles: An Interpretable Multi-vehicle Decision-making and Planning Framework

Wen¹,

Cai²,

Fu³

et al. 2023

Preprint

View full text Add to dashboard Cite

With the development of autonomous driving, it is becoming increasingly common for autonomous vehicles (AVs) and humandriven vehicles (HVs) to travel on the same roads. Existing singlevehicle planning algorithms on board struggle to handle sophisticated social interactions in the real world. Decisions made by these methods are difficult to understand for humans, raising the risk of crashes and making them unlikely to be applied in practice. Moreover, vehicle flows produced by open-source traffic simulators suffer from being overly conservative and lacking behavioral diversity. We propose a hierarchical multi-vehicle decision-making and planning framework with several advantages. The framework jointly makes decisions for all vehicles within the flow and reacts promptly to the dynamic environment through a high-frequency planning module. The decision module produces interpretable action sequences that can explicitly communicate self-intent to the surrounding HVs. We also present the cooperation factor and trajectory weight set, bringing diversity to autonomous vehicles in traffic at both the social and individual levels. The superiority of our proposed framework is validated through experiments with multiple scenarios, and the diverse behaviors in the generated vehicle trajectories are demonstrated through closed-loop simulations.

show abstract

M2I: From Factored Marginal Trajectory Prediction to Interactive Prediction

Cited by 50 publications

References 23 publications

Context-Aware Timewise VAEs for Real-Time Vehicle Trajectory Prediction

Context-Aware Timewise VAEs for Real-Time Vehicle Trajectory Prediction

Beyond RMSE: Do Machine-Learned Models of Road User Interaction Produce Human-Like Behavior?

Bringing Diversity to Autonomous Vehicles: An Interpretable Multi-vehicle Decision-making and Planning Framework

Contact Info

Product

Resources

About