BiTraP: Bi-directional Pedestrian Trajectory Prediction with Multi-modal Goal Estimation

29 Jul 2020  ·  Yu Yao, Ella Atkins, Matthew Johnson-Roberson, Ram Vasudevan, Xiaoxiao Du ·

Pedestrian trajectory prediction is an essential task in robotic applications such as autonomous driving and robot navigation. State-of-the-art trajectory predictors use a conditional variational autoencoder (CVAE) with recurrent neural networks (RNNs) to encode observed trajectories and decode multi-modal future trajectories... This process can suffer from accumulated errors over long prediction horizons (>=2 seconds). This paper presents BiTraP, a goal-conditioned bi-directional multi-modal trajectory prediction method based on the CVAE. BiTraP estimates the goal (end-point) of trajectories and introduces a novel bi-directional decoder to improve longer-term trajectory prediction accuracy. Extensive experiments show that BiTraP generalizes to both first-person view (FPV) and bird's-eye view (BEV) scenarios and outperforms state-of-the-art results by ~10-50%. We also show that different choices of non-parametric versus parametric target models in the CVAE directly influence the predicted multi-modal trajectory distributions. These results provide guidance on trajectory predictor design for robotic applications such as collision avoidance and navigation systems. read more

PDF Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Multi-future Trajectory Prediction JAAD Bitrap-GMM MSE(0.5) 153 # 3
MSE(1.0) 250 # 3
MSE(1.5) 585 # 3
C_MSE(1.5) 501 # 3
CF_MSE(1.5) 998 # 3
Multi-future Trajectory Prediction JAAD Bitrap-NP MSE(0.5) 38 # 2
MSE(1.0) 94 # 2
MSE(1.5) 222 # 2
C_MSE(1.5) 177 # 2
CF_MSE(1.5) 565 # 2
Trajectory Prediction JAAD BiTrap-D MSE(0.5) 93 # 2
MSE(1.0) 378 # 2
MSE(1.5) 1206 # 2
C_MSE(1.5) 1105 # 2
CF_MSE(1.5) 4565 # 2
Trajectory Prediction PIE Bitrap-D MSE(0.5) 41 # 2
MSE(1.0) 161 # 2
MSE(1.5) 511 # 2
C_MSE(1.5) 481 # 2
CF_MSE(1.5) 1949 # 2
Multi-future Trajectory Prediction PIE BiTrap-GMM MSE(0.5) 38 # 3
MSE(1.0) 90 # 3
MSE(1.5) 209 # 3
C_MSE(1.5) 171 # 3
CF_MSE(1.5) 368 # 3
Multi-future Trajectory Prediction PIE BiTrap-NP MSE(0.5) 23 # 2
MSE(1.0) 48 # 2
MSE(1.5) 102 # 2
C_MSE(1.5) 81 # 2
CF_MSE(1.5) 261 # 2

Methods