Planning-based prediction for pedestrians

By Brian D. Ziebart, Nathan D. Ratliff, Garratt Gallagher, Christoph Mertz, K. Peterson, J. Bagnell, Martial Hebert, Anind K. Dey, Siddhartha S. Srinivasa. 2009 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2009

544 citations recorded by the source corpus.

We present a novel approach for determining robot movements that efficiently accomplish the robot's tasks while not hindering the movements of people within the environment. Our approach models the goal-directed trajectories of pedestrians using maximum entropy inverse optimal control. The advantage of this modeling approach is the generality of its learned cost function to changes in the environment and to entirely different environments. We employ the predictions of this model of pedestrian trajectories in a novel incremental planner and quantitatively show the improvement in hindrance-sensitive robot trajectory planning provided by our approach.

Subfields: Automotive Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/36fb553aa996885017afe3489a8377eceddc08ee

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Automotive Engineering

In Computer Vision and Pattern Recognition

Citations in the corpus

Socially compliant mobile robot navigation via inverse reinforcement learning (2016, background): “Taking into account the parts of the pedestrians’ trajectories that the robot has observed so far, similar to Ziebart et al. (2009), may further improve the predictions.”
CHOMP: Covariant Hamiltonian optimization for motion planning (2013, background): “Modeling distributions over a space of paths is a problem of fundamental signiﬁcance to a number of ﬁelds such as statistical physics [10], ﬁnance [45], human behavior modeling [81, 82], and robust and stochastic control [2, 74, 76].”
Core Challenges of Social Robot Navigation: A Survey (2021, methodology): “[262] employ inverse reinforcement learning as a technique to recover models of human navigation objectives and use them to design motion planners for humanlike robot navigation.”
Shared autonomy via hindsight optimization for teleoperation and teaming (2017, methodology): “…of all state-action pairs, dynamic programming can be utilized to compute this through soft-minimum value iteration when the state is discrete (Ziebart et al., 2009, 2012): ∼ · | The log partition function is given by the soft value function, ) (cid:1) , where the integral is over all…”
TNT: Target-driveN Trajectory Prediction (2020, methodology): “[28] proposed planning-based prediction for pedestrians, they first estimated a Bayesian posterior distribution of destinations, and then used inverse reinforcement learning (IRL) to plan the trajectories.”
First-Person Activity Forecasting with Online Inverse Reinforcement Learning (2016, methodology): “IRL has been used to model taxi driver behavior [34] and pedestrian behavior [35, 7].”
A Survey of Inverse Reinforcement Learning: Challenges, Methods and Progress (2018, background): “In this use, applications include destination or route prediction for taxis [13, 14], footstep prediction for planning legged locomotion [15], anticipation of pedestrian interactions [16], energy efficient driving [17], and penetration of a perimeter patrol [18] by learning the patrollers’ preferenc”
Who are you with and where are you going? (2011, background): “The pedestrian behavior model has been extensively studied in the ﬁelds where simulation plays an important role, such as graphics [5], and civil engineering [10] [11], or where accurate prediction is required, such as robotics [13, 6].”
Efficient Exploration via State Marginal Matching (2019, background): “The second term p(z | s) is an estimate of which goal the agent is trying to reach, similar to objectives in intent inference (Ziebart et al., 2009; Xie et al., 2013).”
Social LSTM: Human Trajectory Prediction in Crowded Spaces (2016, unknown): “[78, 23] presented a planning based approach.”
Learning strategies in table tennis using inverse reinforcement learning (2014, background): “2008),modeling goal-directed trajectories of pedestrians (Ziebart et al. 2009) and user simulation in spoken dialog management systems (Chandramohan et al.”
Information gathering actions over human internal state (2016, background): “Work in robotics and perception has focused thus far on estimating these internal states by providing algorithms with observations of humans acting, be it intent prediction [23], [13], [6], [3], [4], [16], Inverse Reinforcement Learning [1], [12], [15], [22], [18], driver style prediction [11], affe”
Exploring applications of deep reinforcement learning for real-world autonomous driving systems (2019, background): “Authors in (Ziebart et al., 2009) perform maximum entropy inverse optimal control to learn a generic cost function for a robot to avoid pedestrians.”
Human Action Recognition and Prediction: A Survey (2018, methodology): “[194] presented a planning based approach for trajectory prediction.”
Teleoperation with intelligent and customizable interfaces (2013, background): “The changes propagate in the same way to the rest of the space, independently of the nature of the change, as in Fig.10(left).”
Human-robot collaborative manipulation planning using early prediction of human motion (2013, background): “Similar to our approach, researchers have explored how to learn models of human motion and use them for online online navigation planning [12].”
It Is Not the Journey but the Destination: Endpoint Conditioned Trajectory Prediction (2020, background): “Rather, humans have the will to exert causal forces to change their motion and constantly adjust their paths as they navigate around obstacles to achieve their goals [4].”
Shared Autonomy via Hindsight Optimization (2015, methodology): “They then derive a method for inferring a distribution over goals from user inputs, where probabilities correspond to how efﬁciently the inputs achieve each goal [29].”