View on GitHub

RISE Labs - HRI Research Toolkit

add description in _config.yml

Return Home Page

Introduction

Driver error factors in 94% of causes of car crashes. Self-driving vehicles help in reduced such driver error saving lives. Industrial interests in autonomous vehicles is of growing interest with automated deliveries amid safety interests during Covid-19 pandemic. Autonomous Vehicles need to be safe as well as should be able to interact with other vehicles in crowded environment. Often times data driven systems fail to work in previously unseen environment. We provide here some of the algorithms AV interactions have been modeled with some of the available datasets.

Data Driven Multi Agent Trajectory Prediction

With the inception of deep network, Autoencoders and (Generative Adversarial Network) GAN have shown great performance in computer vision applications. However when it comes to encoding interactions in AV setup, the temporal and topological contextual information for all the participating agents are imperative. Some of the recent data driven network encoding such information are presented below:

Social GAN

One of the most sort out paper, the proposed GAN model was encoded each agent’s motion in interpersonal level with the agents in the environment with social acceptance clause and further made the system multimodal. The presented social pooling algorithm has influenced other GAN based model for predicting multi agent trajectory.

CGNS

the authors proposed Conditional Generative Neural Systems (CGNS) which jointly predict future trajectories of multiple highly-interactive agents, which takes into account the static context information, interactions among multiple entities and feasibility constraints. A block attention mechanism and a Gaussian mixture attention mask are proposed and applied to historical trajectories and scene image sequences respectively, which are computationally efficient. Hence the proposed system is context aware, interaction aware, feasibility aware and provides probabilistic condition on predicted algorithm.

Social-WaGDAT

The authors propose a generic generative neural system (called Social-WaGDAT) for multi-agent trajectory prediction where explicit interaction are modeling by incorporating relational inductive biases with a dynamic graph representation and leverages both trajectory and scene context information. They further employ an efficient kinematic constraint layer applied to vehicle trajectory prediction which not only ensures physical feasibility but also enhances model performance. The system has been evaluated on three public benchmark datasets for trajectory prediction, where the agents cover pedestrians, cyclists and on-road vehicles. The ablation and baseline study carried out in the paper gives an outline on the state of the art algorithms in such social models.

Probabilistic Crowd GAN

Probabilistic Crowd GAN, extends recent work in trajectory prediction, combining Recurrent Neural Networks (RNNs) with Mixture Density Networks (MDNs) to output probabilistic multimodal predictions, from which likely modal paths are found and used for adversarial training. the use of Graph Vehicle- Pedestrian Attention Network (GVAT), which models social interactions and allows input of a shared vehicle feature, showing that inclusion of this module leads to improved trajectory prediction both with and without the presence of a vehicle.

Game Theoretic Trajectory Prediction

Unlike the data driven counterpart, the game theoretic models represents the agents with strategies to reduce their cost function while in interaction with the other agents. Nash Equilibrium solutions are heavily exploited in such interactions. Depending on the discrete or continuous action and state space, the solutions to the interaction can be based off of graphical solution to solving Hamilton Jacobi Issac’s solution. Below are some of the papers that have used game theoretic trajectory predictions. In some solutions the authors deal with incomplete information setting where the intent of the other agents are not present. Here intent refers to the goal directed behavior which is part of any optimization based algorithm.

Safe and Interactive Autonomy

The presented work is the thesis work of Dorsa Sadign who presented a comprehensive study for Nash Equilibrium solutions in multi agent interaction. The driving interaction in belief update and preferences are highlighted which provides different motion. The papers with the subsequent codes are available in Dorsa’s github here.

Courteous Autonomous Cars

Besides performing a game theoretic solution for multi agent interactions, the authors here formalize courtesy as a term in the objective that measures the increase in another driver’s cost induced by the autonomous car’s behavior. The courteous behavior aids in explaining human driving. The authors have further validated the proposed algorithm in NSGIM Dataset.

Social Behavior in Autonomous Vehicles

We present a framework that integrates social psychology tools into controller design for autonomous vehicles. Our key insight utilizes Social Value Orientation (SVO), quantifying an agent’s degree of selfishness or altruism, which allows us to better predict driver behavior. The presented algorithm has been further validated in the NSGIM dataset

Available Datasets

Sample Projects

Additional Videos and Media

Additional Reading

Return Home Page