Regarding the Issue of Human Trajectory Prediction

[paofuu]

I would like to start by expressing my gratitude for the significant contributions you and your team have made to the field of dynamic obstacle avoidance. I am currently conducting research in this area myself.
I noticed that in your 2023 paper, "Intention Aware Robot Crowd Navigation with Attention-Based Interaction Graph," your method incorporated human trajectory prediction as an input for attention calculation. However, I observed that in your 2025 project (and the corresponding paper), trajectory prediction does not seem to be included in the input.I was wondering if you could share the rationale behind this change? Did specific experimental results or other considerations lead you to move away from that approach?
I understand you are very busy, but I would be extremely grateful for any insights you could share.

[Shuijing725]

Thank you for your interest and your question! From what I remember, we tried adding the ground truth future trajectories to the observation space, like what the 2023 paper did, but did not observe too much performance gain. The reason might be human-human interactions are sparser due to occlusions caused by static obstacles in our new simulation environment.
In addition, predicting human trajectories in the real-world requires tracking the IDs of humans, which increases the computation load of our system and is prone to perception noises.