Social NCE: Contrastive Learning of Socially-aware Motion Representations (2020-12-21T00:00:00.000000Z)

TL;DR

This work intro-duce a social contrastive loss that regularizes the extracted motion representation by discerning the ground-truth positive events from synthetic negative ones, and constructs informative negative samples based on the prior knowledge of rare but dangerous circumstances.

Abstract

Learning socially-aware motion representations is at the core of recent advances in multi-agent problems, such as human motion forecasting and robot navigation in crowds. Despite promising progress, existing representations learned with neural networks still struggle to generalize in closed-loop predictions (e.g., output colliding trajectories). This issue largely arises from the non-i.i.d. nature of sequential prediction in conjunction with ill-distributed training data. Intuitively, if the training data only comes from human behaviors in safe spaces, i.e., from "positive" examples, it is difficult for learning algorithms to capture the notion of "negative" examples like collisions. In this work, we aim to address this issue by explicitly modeling negative examples through self-supervision: (i) we intro-duce a social contrastive loss that regularizes the extracted motion representation by discerning the ground-truth positive events from synthetic negative ones; (ii) we construct informative negative samples based on our prior knowledge of rare but dangerous circumstances. Our method substantially reduces the collision rates of recent trajectory forecasting, behavioral cloning and reinforcement learning algorithms, outperforming state-of-the-art methods on several benchmarks. Our code is available at https://github.com/vita-epfl/social-nce.

Authors

Alexandre Alahi

17 papers

Qi Yan

2 papers

Yuejiang Liu

4 papers

TL;DR

Abstract

Authors

References107 items

Interpretable Social Anchors for Human Trajectory Forecasting in Crowds

Injecting Knowledge in Data-driven Vehicle Trajectory Predictors

Negative Data Augmentation

Exploring Simple Siamese Representation Learning

DROGON: A Trajectory Prediction Model based on Intention-Conditioned Behavior Reasoning

Contrastive Learning with Hard Negative Samples

Hard Negative Mixing for Contrastive Learning

End-to-end Contextual Perception and Prediction with Interaction Transformer

DSDNet: Deep Structured self-Driving Network

Contrastive Learning for Unpaired Image-to-Image Translation

Demystifying Contrastive Self-Supervised Learning: Invariances, Augmentations and Dataset Biases

Multi-label Contrastive Predictive Coding

Human Trajectory Forecasting in Crowds: A Deep Learning Perspective

Debiased Contrastive Learning

Space-Time Correspondence as a Contrastive Random Walk

Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems

Disagreement-Regularized Imitation Learning

Supervised Contrastive Learning

Recursive Social Behavior Graph for Trajectory Prediction

EvolveGraph: Multi-Agent Trajectory Prediction with Dynamic Relational Reasoning

Plato

Social-STGCNN: A Social Spatio-Temporal Graph Convolutional Neural Network for Human Trajectory Prediction

A Simple Framework for Contrastive Learning of Visual Representations

Multi-Agent Motion Planning for Dense and Dynamic Environments via Deep Reinforcement Learning

Trajectron++: Dynamically-Feasible Trajectory Forecasting with Heterogeneous Data

Self-Supervised Learning of Pretext-Invariant Representations

Off-Policy Policy Gradient Algorithms by Constraining the State Distribution Shift

Momentum Contrast for Unsupervised Visual Representation Learning

MultiPath: Multiple Probabilistic Anchor Trajectory Hypotheses for Behavior Prediction

STGAT: Modeling Spatial-Temporal Interactions for Human Trajectory Prediction

Learning Self-Correctable Policies and Value Functions from Demonstrations with Negative Sampling

An Optimistic Perspective on Offline Reinforcement Learning

Social-BiGAT: Multimodal Trajectory Forecasting using Bicycle-GAN and Graph Attention Networks

Stabilizing Off-Policy Q-Learning via Bootstrapping Error Reduction

Causal Confusion in Imitation Learning

SQIL: Imitation Learning via Reinforcement Learning with Sparse Rewards

Random Expert Distillation: Imitation Learning via Expert Policy Support Estimation

Human motion trajectory prediction: a survey

PRECOG: PREdiction Conditioned on Goals in Visual Multi-Agent Settings

Challenges of Real-World Reinforcement Learning

Social Ways: Learning Multi-Modal Distributions of Pedestrian Trajectories With GANs

Exploring the Limitations of Behavior Cloning for Autonomous Driving

Benchmarking Neural Network Robustness to Common Corruptions and Perturbations

A Theoretical Analysis of Contrastive Unsupervised Representation Learning

Collaborative Sampling in Generative Adversarial Networks

Off-Policy Deep Reinforcement Learning without Exploration

ChauffeurNet: Learning to Drive by Imitating the Best and Synthesizing the Worst

A Literature Review on the Prediction of Pedestrian Behavior in Urban Scenarios

The Trajectron: Probabilistic Multi-Agent Trajectory Modeling With Dynamic Spatiotemporal Graphs

Map-based Deep Imitation Learning for Obstacle Avoidance

Discriminator Rejection Sampling

Crowd-Robot Interaction: Crowd-Aware Robot Navigation With Attention-Based Deep Reinforcement Learning

Discriminator-Actor-Critic: Addressing Sample Inefficiency and Reward Bias in Adversarial Imitation Learning

Representation Learning with Contrastive Predictive Coding

SoPhie: An Attentive GAN for Predicting Paths Compliant to Social and Physical Constraints

Convolutional Social Pooling for Vehicle Trajectory Prediction

Unsupervised Feature Learning via Non-parametric Instance Discrimination

Social GAN: Socially Acceptable Trajectories with Generative Adversarial Networks

An efficient framework for learning sentence representations

Deep Contextualized Word Representations

WILDTRACK: A Multi-camera HD Dataset for Dense Unscripted Pedestrian Detection

Social Attention: Modeling Attention in Human Crowds

Rainbow: Combining Improvements in Deep Reinforcement Learning

3DOF Pedestrian Trajectory Prediction Learned from Long-Term Autonomous Mobile Robot Deployment Data

Time-Contrastive Networks: Self-Supervised Learning from Video

DESIRE: Distant Future Prediction in Dynamic Scenes with Interacting Agents

DART: Noise Injection for Robust Imitation Learning

Socially aware motion planning with deep reinforcement learning

Unsupervised Learning of Sentence Embeddings Using Compositional n-Gram Features

Deeply AggreVaTeD: Differentiable Imitation Learning for Sequential Prediction

Improved Deep Metric Learning with Multi-class N-pair Loss Objective

Decentralized non-communicating multiagent collision avoidance with deep reinforcement learning

Point-based path prediction from polar histograms

Social LSTM: Human Trajectory Prediction in Crowded Spaces

Generative Adversarial Imitation Learning

Adam: A Method for Stochastic Optimization