Improved Activity Forecasting for Generating Trajectories

TL;DR

This paper introduces a novel inverse reinforcement learning approach that combines 2D and 3D activity forecasting with convolutional value iteration, resulting in more accurate and faster trajectory generation, especially demonstrated on seabird data.

Contribution

It proposes a new convolutional value iteration method and modifies the reward function with an Lp norm for improved trajectory forecasting.

Findings

Outperforms previous methods in MHD error

Faster trajectory generation

Produces trajectories closely matching real seabird paths

Abstract

An efficient inverse reinforcement learning for generating trajectories is proposed based of 2D and 3D activity forecasting. We modify reward function with $L_p$ norm and propose convolution into value iteration steps, which is called convolutional value iteration. Experimental results with seabird trajectories (43 for training and 10 for test), our method is best in terms of MHD error and performs fastest. Generated trajectories for interpolating missing parts of trajectories look much similar to real seabird trajectories than those by the previous works.