Human Motion Prediction Under Social Grouping Constraints

Abstract

Accurate long-term prediction of human motion in populated spaces is an important but difficult task for mobile robots and intelligent vehicles. What makes this task challenging is that human motion is influenced by a large variety of factors including the person's intention, the presence, attributes, actions, social relations and social norms of other surrounding agents, and the geometry and semantics of the environment. In this paper, we consider the problem of computing human motion predictions that account for such factors. We formulate the task as an MDP planning problem with stochastic policies and propose a weighted random walk algorithm in which each agent is locally influenced by social forces from other nearby agents. The novelty of this paper is that we incorporate social grouping information into the prediction process reflecting the soft formation constraints that groups typically impose to their members' motion. We show that our method makes more accurate predictions than three state-of-the-art methods in terms of probabilistic and geometrical performance metrics.

Human motion trajectory prediction: a survey

“…obstacle-aware methods , which account for the presence of individual static obstacles (e.g., Alahi et al, 2016; Althoff et al, 2008b; Bera et al, 2016; Elfring et al, 2014; Ferrer and Sanfeliu, 2014; Rehder and Klöden, 2015; Trautman and Krause, 2010; Vemula et al, 2017); 3.3. map-aware methods , which account for environment geometry and topology (e.g., Chen et al, 2017; Chung and Huang, 2010, 2012; Gong et al, 2011; Henry et al, 2010; Ikeda et al, 2012; Kooij et al, 2019; Liao et al, 2003; Pfeiffer et al, 2016; Pool et al, 2017; Rösmann et al, 2017; Rudenko et al, 2017, 2018b; Vasquez, 2016; Yen et al, 2008; Ziebart et al, 2009); 3.4. semantics-aware methods , which additionally account for environment semantics or affordances such as no-go zones, crosswalks, sidewalks, or traffic lights (e.g., Ballan et al, 2016; Coscia et al, 2018; Karasev et al, 2016; Kitani et al, 2012; Kuhnt et al, 2016; Lee et al, 2017; Ma et al, 2017; Rehder et al, 2018; Zheng et al, 2016).…”

“…3.3. map-aware methods , which account for environment geometry and topology (e.g., Chen et al, 2017; Chung and Huang, 2010, 2012; Gong et al, 2011; Henry et al, 2010; Ikeda et al, 2012; Kooij et al, 2019; Liao et al, 2003; Pfeiffer et al, 2016; Pool et al, 2017; Rösmann et al, 2017; Rudenko et al, 2017, 2018b; Vasquez, 2016; Yen et al, 2008; Ziebart et al, 2009);…”

“…Rudenko et al (2018a) extended the MDP-based approaches (Karasev et al, 2016; Ziebart et al, 2009) with a fast random-walk-based method to generate joint predictions for all observed people using social forces. The authors extended their approach considering group-based social motion constraints in Rudenko et al (2018b). Wu et al (2018) extended the gridmap transition-based and reachability-based framework (Coscia et al, 2018; Rehder and Klöden, 2015) with automatic inference of local goal points, and calculate the stochastic policy in each cell, augmenting the physics-based dynamics with optimal motion direction.…”

“…Most essential cues, used to predict future states of an agent, are related to the agent itself. To this end most of the algorithms use current position and velocity of the target agent (Bahram et al, 2016; Bansal et al, 2019; Bennewitz et al, 2005; Bera et al, 2016; Elfring et al, 2014; Ferrer and Sanfeliu, 2014; Habibi et al, 2018; Karasev et al, 2016; Kitani et al, 2012; Kucner et al, 2017; Kuderer et al, 2012; Luo and Cai, 2019; Pellegrini et al, 2009; Rhinehart et al, 2018a; Rudenko et al, 2018b; Trautman and Krause, 2010; Wu et al, 2018; Ziebart et al, 2009), often considering also the history of recent states/velocities. Position and velocity are also the main attributes of the target agent in vehicle motion prediction tasks (Broadhurst et al, 2005; Hermes et al, 2009; Käfer et al, 2010).…”

THÖR: Human-Robot Navigation Data Collection and Accurate Motion Trajectories Dataset

“…In addition to basic trajectory data, state-of-the-art methods for tracking or motion prediction, for instance, can also incorporate information about the environment, social grouping, head orientation or personal traits. For instance, Lau et al [13] estimate social grouping formations during tracking and Rudenko et al [20] use group affiliation as a contextual cue to predict future motion. Unhelkar et al [24] use head orientation to disambiguate and recognize typical motion patterns that people are following.…”