ICRA2024
MacFormer: Map-Agent Coupled Transformer for Real-time and Robust Trajectory Prediction. [arXiv]
ICCV2023
ICCV 2023 All Paper [List]
ADAPT: Efficient Multi-Agent Trajectory Prediction with Adaptation. [Paper][arXiv][Website]
BeLFusion: Latent Diffusion for Behavior-Driven Human Motion Prediction. [Paper][arxiv][Code][Website]
BiFF: Bi-level Future Fusion with Polyline-based Coordinate for Interactive Trajectory Prediction. [Paper][arXiv]
EigenTrajectory: Low-Rank Descriptors for Multi-Modal Trajectory Forecasting. [Paper][arXiv]
Fast Inference and Update of Probabilistic Density Estimation on Trajectory Prediction. [Paper][arxiv][Code]
Forecast-MAE: Self-supervised Pre-training for Motion Forecasting with Masked Autoencoders. [Paper][arXiv][Code]
GameFormer: Game-theoretic Modeling and Learning of Transformer-based Interactive Prediction and Planning for Autonomous Driving. [Paper][arXiv][Code][Website]
HumanMAC: Masked Motion Completion for Human Motion Prediction. [Paper][arxiv][Code][Website]
INT2: Interactive Trajectory Prediction at Intersections. [Paper][Code][Website]
InterDiff: Generating 3D Human-Object Interactions with Physics-Informed Diffusion. [Paper][arxiv][Code][Website]
Joint Metrics Matter: A Better Standard for Trajectory Forecasting. [Paper][arXiv]
Learn TAROT with MENTOR: A Meta-Learned Self-Supervised Approach for Trajectory Prediction. [Paper]
R-Pred: Two-Stage Motion Prediction Via Tube-Query Attention-Based Trajectory Refinement. [Paper][arXiv]
Semi-supervised Semantics-guided Adversarial Training for Robust Trajectory Prediction. [Paper][arXiv]
Sparse Instance Conditioned Multimodal Trajectory Prediction. [Paper]
Traj-MAE: Masked Autoencoders for Trajectory Prediction. [Paper][arXiv]
Trajectory Unified Transformer for Pedestrian Trajectory Prediction. [Paper]
TrajPAC: Towards Robustness Verification of Pedestrian Trajectory Prediction Models. [Paper][arxiv]
CVPR2023
CVPR 2023 All Paper [List]
Decompose More and Aggregate Better: Two Closer Looks at Frequency Representation Learning for Human Motion Prediction. [Paper]
DeFeeNet: Consecutive 3D Human Motion Prediction with Deviation Feedback. [Paper][arXiv]
EqMotion: Equivariant Multi-agent Motion Prediction with Invariant Interaction Reasoning. [Paper][arXiv][Code]
FEND: A Future Enhanced Distribution-Aware Contrastive Learning Framework for Long-tail Trajectory Prediction. [Paper][arXiv]
FJMP: Factorized Joint Multi-Agent Motion Prediction over Learned Directed Acyclic Interaction Graphs. [Paper][arXiv][Code]
IPCC-TP: Utilizing Incremental Pearson Correlation Coefficient for Joint Multi-Agent Trajectory Prediction. [Paper][arXiv]
Leapfrog Diffusion Model for Stochastic Trajectory Prediction. [Paper][arXiv][Code]
MotionDiffuser: Controllable Multi-Agent Motion Prediction using Diffusion. [Paper]
Planning-oriented Autonomous Driving. [Paper][arXiv][Code][Website]
ProphNet: Efficient Agent-Centric Motion Forecasting with Anchor-Informed Proposals. [Paper][arXiv]
Query-Centric Trajectory Prediction. [Paper]
Stimulus Verification is a Universal and Effective Sampler in Multi-modal Human Trajectory Prediction. [Paper]
Trace and Pace: Controllable Pedestrian Animation via Guided Trajectory Diffusion. [Paper][arXiv][Website]
Trajectory-Aware Body Interaction Transformer for Multi-Person Pose Forecasting. [Paper][arXiv][Code]
Uncovering the Missing Pattern: Unified Framework Towards Trajectory Imputation and Prediction. [Paper][Code]
Unsupervised Sampling Promoting for Stochastic Human Trajectory Prediction. [Paper]
ViP3D: End-to-end Visual Trajectory Prediction via 3D Agent Queries. [Paper][arXiv][Website]
Weakly Supervised Class-agnostic Motion Prediction for Autonomous Driving. [Paper]
ICLR2023
ICLR 2023 All Paper [List]
- Learning Heterogeneous Interaction Strengths by Trajectory Prediction with Graph Neural Network. [OpenReview][arXiv]
https://github.com/nokpil/RAIN 主要是运动中的人，球等的轨迹预测。【2023.11.20/Monday】 - Stochastic Multi-Person 3D Motion Forecasting. [OpenReview]
本文旨在解决以往人类运动预测工作中忽略的现实世界复杂性，强调多人运动的社会属性、运动和社会互动的多样性以及关节运动的复杂性。为此，我们引入了一种新的任务——随机多人三维运动预测。我们提出了一个双层生成建模框架，分别在局部水平上对独立的个人运动和在全局水平上对社会互动进行建模。值得注意的是，这种双层建模机制可以在共享生成模型中实现，通过引入代表未来运动意图的可学习潜在代码，并在不同级别切换代码的操作模式。我们的框架是通用的;我们用不同的生成模型来实例化它，包括生成对抗网络和扩散模型，以及各种多人预测模型。在mu – mocap, MuPoTS-3D和SoMoF基准上进行的广泛实验表明，我们的方法产生了多样化和准确的多人预测，显着优于最先进的状态。【2023.11.20/Monday】ICRA2023
- Expanding the Deployment Envelope of Behavior Prediction via Adaptive Meta-Learning. [arXiv]
这个主要是基于不同数据集作迁移学习的，这两篇都是来自英伟达同一个作者，实验基于第一篇的unified dataset ：https://github.com/NVlabs/trajdata， 为各种开源轨迹预测数据集提供了统一接口。在单个方法上可能没有什么创新性，主要是结合了元学习，数学性比较强。【2023.11.20/Monday】
《归纳偏置/归纳偏差/inductive bias》
“归纳偏差”中的“偏差”容易让人想到数据估计中估计值与真实值的差别error，会让人认为“归纳偏差”是某种需要被规避的错误、误差，但事实并非如此，inductive bias在机器学习中起到的是积极作用。
更合适的翻译应该是“归纳偏置”：归纳是自然科学中常用的两大方法（归纳与演绎，induction and deduction)之一,指的是从一些例子中寻找共性、泛化，形成一个比较通用的规则的过程；而“bias”“偏置”是指我们对模型的偏好。因此归纳偏置可以理解为，从现实生活中观察到的现象中归纳出一定的规则，然后对模型做一定的约束，从而可以起到“模型选择”的作用，即从假设空间中选择出更符合现实规则的模型。类似贝叶斯学习中的“先验，prior”，但与“先验”不同的是，“先验”一般是确定性的已知、知识，而“偏置”更倾向于是一种经验性的假设。
归纳偏置在机器学习中几乎无处不在。具体的例子可以参考文末资料，笔者说一下自己想到的一个简单例子。
比如，我们考虑一组（0，0）（1，1）…（i，i）…（n,n）的点，当我们要学习一个模型来模拟x到y的映射时，我们最希望的是模型学习到y=x这个线性映射，可实际上符合这些样本点的映射有无穷多种，如果我们在非线性映射空间进行学习的话，学习过程就会十分复杂，因此我们在某些情况下就会做出“我们假定这是一个线性映射”这样的假设，之后在此假设的基础上对模型进行学习，这里的“我们假定这是一个线性映射”就是基于先验知识等所作出的归纳偏置；而当我们在线性映射空间进行学习时，也有可能得到无限种映射，此时我们会根据奥卡姆剃刀原则选择“尽可能简单的模型”。奥卡姆剃刀是机器学习中最典型的一种归纳偏置。
作用：
机器学习中的归纳偏置可以提高模型的泛化性。例如在上文的例子中，如果我们在采样中遇到了噪音（0，10），（7，-20）…，等就很可能会使得对泛化性更强的映射y=x学习的失败，从而学习到一个“过拟合”的模型，而在我们加入“线性映射”、“奥卡姆剃刀”等归纳偏置后，就会更容易学习到在目标域更具有泛化性、通用性的映射y=x（模型）
总结：
inductive bias更合适的翻译是归纳偏置而非归纳偏差，它是一种在模型的无限解空间中所引入的合理假设与约束，这类假设、约束能够缩小求解空间并提高所得模型在目标域的泛化性。
arXiv2022
Safety-compliant Generative Adversarial Networks for Human Trajectory Forecasting. [arXiv]
Wayformer: Motion Forecasting via Simple & Efficient Attention Networks. [arXiv]
CoRL2022
SSL-Lanes: Self-Supervised Learning for Motion Forecasting in Autonomous Driving. [arXiv][Code]
NeurIPS2022
NeurIPS 2022 All Paper [List]
Contact-aware Human Motion Forecasting. [OpenReview][arXiv][Code]
Forecasting Human Trajectory from Scene History. [OpenReview][arXiv]
Motion Forecasting Transformer with Global Intention Localization and Local Movement Refinement. [OpenReview][arXiv][Code]
ECCV2022
ECCV 2022 All Paper [List]
Action-based Contrastive Learning for Trajectory Prediction. [arXiv]
AdvDO: Realistic Adversarial Attacks for Trajectory Prediction. [arXiv]
Aware of the History: Trajectory Forecasting with the Local Behavior Data. [arXiv][Code]
Diverse Human Motion Prediction Guided by Multi-Level Spatial-Temporal Anchors. [arXiv] [Code][Website]
D2-TPred: Discontinuous Dependency for Trajectory Prediction under Traffic Lights. [arXiv][Code]
Entry-Flipped Transformer for Inference and Prediction of Participant Behavior. [arXiv]
Hierarchical Latent Structure for Multi-Modal Vehicle Trajectory Forecasting. [arXiv]
Human Trajectory Prediction via Neural Social Physics. [arXiv][Code]
Learning Pedestrian Group Representations for Multi-modal Trajectory Prediction. [Website][arXiv][Code]
Overlooked Poses Actually Make Sense: Distilling Privileged Knowledge for Human Motion Prediction. [arXiv]
Polarimetric Pose Prediction. [arXiv]
– PreTraM: Self-Supervised Pre-training via Connecting Trajectory and Map. [arXiv][Code]
Skeleton-Parted Graph Scattering Networks for 3D Human Motion Prediction. [arXiv][Code]
Social-Implicit: Rethinking Trajectory Prediction Evaluation and The Effectiveness of Implicit Maximum Likelihood Estimatio. [Website][arXiv][Code]
Social ODE: Multi-Agent Trajectory Forecasting with Neural Ordinary Differential Equations.
SocialVAE: Human Trajectory Prediction using Timewise Latents. [arXiv][Code]
ST-P3: End-to-end Vision-based Autonomous Driving via Spatial-Temporal Feature Learning. [arXiv][Code]
View Vertically: A Hierarchical Network for Trajectory Prediction via Fourier Spectrums. [arXiv]
CVPR2022
CVPR 2022 轨迹/Motion预测相关论文，Proceedings已出，今年CVPR分外热闹，站了一上午Poster，一上午嘴都没有停，感觉无人车还是热门啊。恐有遗漏，欢迎补充！
先main conference 后workshop 以字母顺序排序
Trajectory Prediction Related
Adaptive Trajectory Prediction via Transferable GNN. [arXiv][Paper]
ATPFL: Automatic Trajectory Prediction Model Design Under Federated Learning Framework. [Paper]
Convolutions for Spatial Interaction Modeling. [arXiv][Paper]
End-to-End Trajectory Distribution Prediction Based on Occupancy Grid Maps. [arXiv][Paper][Code]
Forecasting from LiDAR via Future Object Detection. [arXiv][Paper][Code]
Graph-based Spatial Transformer with Memory Replay for Multi-future Pedestrian Trajectory Prediction. [Paper][Code]
GroupNet: Multiscale Hypergraph Neural Networks for Trajectory Prediction with Relational Reasoning. [arXiv][Paper][Code]
How Many Observations are Enough? Knowledge Distillation for Trajectory Forecasting. [arXiv][Paper]
LTP: Lane-Based Trajectory Prediction for Autonomous Driving. [Paper]
M2I: From Factored Marginal Trajectory Prediction to Interactive Prediction. [arXiv][Paper][Code]
Neural Prior for Trajectory Estimation. [Paper][Website]
Non-Probability Sampling Network for Stochastic Human Trajectory Prediction. [arXiv][Paper][Code]
On Adversarial Robustness of Trajectory Prediction for Autonomous Vehicles. [arXiv][Paper][Code]
Remember Intentions: Retrospective-Memory-based Trajectory Prediction. [arXiv][Paper][Code]
ScePT: Scene-consistent, Policy-based Trajectory Predictions for Planning. [Paper][Code]
Stochastic Trajectory Prediction via Motion Indeterminacy Diffusion. [arXiv][Paper][Code]
Vehicle Trajectory Prediction Works, But Not Everywhere. [arXiv][Paper][Code]
Whose Track Is It Anyway? Improving Robustness to Tracking Errors With Affinity-Based Trajectory Prediction. [Paper][Code]
Goal-driven Self-Attentive Recurrent Networks for Trajectory Prediction. [arXiv][Paper] (CVPR’22 Workshop Precognition: Seeing Through the Future)
Importance Is in Your Attention: Agent Importance Prediction for Autonomous Driving. [arXiv][Paper](CVPR’22 Workshop Precognition: Seeing Through the Future)
Motion Prediction Related
BE-STI: Spatial-Temporal Integrated Network for Class-Agnostic Motion Prediction With Bidirectional Enhancement. [Paper][Code]
Forecasting Characteristic 3D Poses of Human Actions. [arXiv][Paper][Code]
HiVT: Hierarchical Vector Transformer for Multi-Agent Motion Prediction. [Paper][Code]
Human Trajectory Prediction With Momentary Observation. [Paper]
MotionAug: Augmentation With Physical Correction for Human Motion Prediction. [arXiv][Paper][Code]
Motron: Multimodal Probabilistic Human Motion Forecasting. [arXiv][Paper][Code]
Multi-Objective Diverse Human Motion Prediction With Knowledge Distillation. [Paper]
Multi-Person Extreme Motion Prediction. [arXiv][Paper][Code]
Progressively Generating Better Initial Guesses Towards Next Stages for High-Quality Human Motion Prediction. [arXiv][Paper][Code]
Spatial-Temporal Gating-Adjacency GCN for Human Motion Prediction. [arXiv][Paper]
Towards Robust and Adaptive Motion Forecasting: A Causal Representation Perspective. [arXiv][Paper][Code]
Weakly-Supervised Action Transition Learning for Stochastic Human Motion Prediction. [arXiv][Paper][Code]
ICLR2022
之前整理了一个链接
D-CODE: Discovering Closed-form ODEs from Observed Trajectories. [Paper][Code]
Latent Variable Sequential Set Transformers For Joint Multi-Agent Motion Prediction. [arXiv][Paper][Code]
ProtoRes: Proto-Residual Network for Pose Authoring via Learned Inverse Kinematics. [arXiv][Paper][Website]
Scene Transformer: A Unified Architecture for Predicting Multiple Agent Trajectories. [arXiv][Paper]
THOMAS: Trajectory Heatmap Output with learned Multi-Agent Sampling. [arXiv][Paper]
You Mostly Walk Alone: Analyzing Feature Attribution in Trajectory Prediction. [arXiv][Paper]
