Context-based cyclist path prediction using Recurrent Neural Networks

Abstract

This paper proposes a Recurrent Neural Network (RNN) for cyclist path prediction to learn the effect of contextual cues on the behavior directly in an end-to-end approach, removing the need for any annotations. The proposed RNN incorporates three distinct contextual cues: one related to actions of the cyclist, one related to the location of the cyclist on the road, and one related to the interaction between the cyclist and the egovehicle. The RNN predicts a Gaussian distribution over the future position of the cyclist one second into the future with a higher accuracy, compared to a current state-of-the-art model that is based on dynamic mode annotations, where our model attains an average prediction error of 33 cm one second into the future.

Human motion trajectory prediction: a survey

“…Recurrent neural networks (RNNs) for sequence learning, and long short-term memory (LSTM) networks in particular, have recently become a widely popular modeling approach for predicting human (Alahi et al, 2016; Bartoli et al, 2018; Sadeghian et al, 2019; Saleh et al, 2018b; Sun et al, 2018; Varshneya and Srinivasaraghavan, 2017; Vemula et al, 2018), vehicle (Altché and de La Fortelle, 2017; Ding et al, 2019; Kim et al, 2017; Park et al, 2018), and cyclist (Pool et al, 2019) motion. Alahi et al (2016) was the first to propose a Social-LSTM model to predict joint trajectories in continuous spaces.…”

“…Considering additional semantic attributes of the target agent may further refine the quality of predictions: gender and age in Ma et al (2017), personality type (Bera et al, 2017), class of the dynamic agent (e.g. a person or a cyclist in pedestrian areas, motorcycle, car, or a truck on a highway) (Altché and de La Fortelle, 2017; Ballan et al, 2016; Coscia et al, 2018), person’s attention and awareness of the robot’s presence in Oli et al (2013), Kooij et al (2019), and Blaiotta (2019), and raised arm as a bending intention indicator for cyclists (Kooij et al, 2019; Pool et al, 2019).…”

“…Many pattern-based (Bennewitz et al, 2005, 2002; Carvalho et al, 2019; Chen et al, 2016, 2008; Goldhammer et al, 2014; Habibi et al, 2018; Han et al, 2019; Hermes et al, 2009; Huynh and Alaghband, 2019; Kim et al, 2017, 2011; Kucner et al, 2017, 2013; Makansi et al, 2019; Makris and Ellis, 2002; Molina et al, 2018; Nikhil and Tran Morris, 2018; Piciarelli et al, 2005; Ridel et al, 2019; Saleh et al, 2018b; Sung et al, 2012; Suraj et al, 2018; Tadokoro et al, 1993; Thompson et al, 2009; Unhelkar et al, 2015; Wang et al, 2016; Xiao et al, 2015; Xue et al, 2019, 2017) and planning-based methods (Gong et al, 2011; Karasev et al, 2016; Kitani et al, 2012; Rhinehart et al, 2018a; Rudenko et al, 2017; Vasquez, 2016; Yen et al, 2008; Ziebart et al, 2009) are unaware predictors, owing to the increase of complexity for conditioning the learned transition patterns or optimal actions on the presence and positions of other agents. Methods for predicting pedestrians crossing behavior (Gu et al, 2016; Keller and Gavrila, 2014; Kooij et al, 2014; Mínguez et al, 2018; Quintero et al, 2014; Roth et al, 2016; Schulz and Stiefelhagen, 2015) and cyclist motion (Pool et al, 2019, 2017; Saleh et al, 2018a; Zernetsch et al, 2016) typically treat each agent individu...…”

“…Owing to the high level of structure in the environment, methods in autonomous driving scenarios extensively use available semantic information, such as street layout and traffic rules (Agamennoni et al, 2012; Chai et al, 2019; Choi and Savarese, 2010; Cui et al, 2019; Djuric et al, 2018; Gu et al, 2016; Hong et al, 2019; Jain et al, 2019; Keller and Gavrila, 2014; Kooij et al, 2014; Kuhnt et al, 2016; Lee et al, 2017; Petrich et al, 2013; Pool et al, 2019, 2017; Srikanth et al, 2019; Xie et al, 2018) or current state of the traffic lights (Gu et al, 2016; Jain et al, 2019; Karasev et al, 2016), also for predicting pedestrian and cyclist motion (Habibi et al, 2018; Kooij et al, 2019; Koschi et al, 2018).…”

Set-Based Prediction of Traffic Participants Considering Occlusions and Traffic Rules

“…b) Probability distribution: To consider that other traffic participants have infinitely many future behaviors, we can compute a probability distribution, e. g., of kinematic variables using dynamic Bayesian networks [51]- [53]. Furthermore, neural networks have been proposed to predict most likely behaviors of vehicles on highways [54], [55], of pedestrians [56], and of cyclists [57]. For pedestrians, also linear quadratic regulator-based models are used [58].…”