Vol.:(0123456789) 1 3 KI - Künstliche Intelligenz https://doi.org/10.1007/s13218-020-00658-7 TECHNICAL CONTRIBUTION ITP: Inverse Trajectory Planning for Human Pose Prediction Pedro A. Peña 1  · Ubbo Visser 1 Received: 10 March 2020 / Accepted: 6 April 2020 © Gesellschaft für Informatik e.V. and Springer-Verlag GmbH Germany, part of Springer Nature 2020 Abstract Tracking and predicting humans in three dimensional space in order to know the location and heading of the human in the environment is a difcult task. Though if solved it will allow a robotic agent to know where it can safely be and navigate the environment without imposing any danger to the human that it is interacting with. We propose a novel probabilistic framework for robotic systems in which multiple models can be fused into a circular probabilitymap to forecast human poses. We developed and implemented the framework and tested it on Toyota’s HSR robot and Waymo Open Dataset. Our experiments show promising results. 1 Introduction and Motivation Robots tracking and inferring human trajectories is essential to operate in human environments, and it is a difcult task because humans are unpredictable [24, 34]. Moreover, there are studies in the literature that show how humans navigate in pedestrian scenarios [12, 18, 30, 31] or one-to-one interactions with a human in an isolated task such as pick and place [8–10, 23]. Probabilistic motion models [2, 5–7, 13–15] have also shown promising results, and inverse reinforcement learning techniques [19, 21, 22, 33, 41] have also been studied to fnd a reward function that models human behavior such as walking or other activities. Deep learning methods have also become popular [1, 16, 25, 32, 37, 38]. Yet, current methods either suf- fer from assumptions made in the models, i.e., heuristics that do well under certain conditions, overhead of preprocessing datasets, or from the inability to generalize, in the case where a model is trained with biases from data and cannot adapt to new situations. Furthermore, the initiative to integrate various models to increase prediction accuracy in real-world scenarios is seldom seen and methods are tested under constrained envi- ronments. Even in cases where models use real-world scenarios as a testbed such as the work from Rhinehart et al. [33], the state space is constrained to some human activities. The limi- tations of these models refect the complexity of human activ- ity, and the human prediction problem requires frameworks that can utilize various models and operate in realistic human environments. Therefore, we propose a framework for robotic systems that fuses data from perception sensors and/or models to infer human poses. The contribution of this method over other methods are: • A generic probabilistic framework, ITP, for human pose prediction for robotic systems that enables the designer to add various models for specifc applications; • Circular probabilitymap: a human-centered data structure that maps high-level features in the environment to a rep- resentation that can be fused for human pose prediction in robotic systems; To the best of our knowledge, this is the frst human forecast framework for robotic systems that allows a designer to add application-specifc models without having to retrain on data. We organize the paper as follows: in Sect. 2, we will dis- cuss existing models for human motion prediction and plan- ners. We formulate the person trajectory prediction as an inverse trajectory planning problem, explained in Sect. 3.1, and introduce a novel probabilistic framework that allows us to tackle the intractability of human pose prediction in Sect. 3.2 including an explanation of the human-specifc This Technical Contribution is part of Pecora, F., Mansouri, M., Hawes N. and Kunze L. (Eds.). Special Issue on Reintegrating Artifcial Intelligence and Robotics. Künstl Intell, Volume 33, Issue 4, December 2019. https://link.springer.com/journal/13218/33/4. * Pedro A. Peña pedro@cs.miami.edu Ubbo Visser visser@cs.miami.edu 1 Department of Computer Science, University of Miami, Coral Gables, FL 33146, USA