Open fcjian opened 3 months ago
still cleaning the code. Will open source tomorrow.
Thank you. I have another question. For the experiment in Table 5 of the paper, how can high-level commands be integrated into OmniDrive? And how can Ego Status be incorporated into OmniDrive for both BEV and Planner?
Planner:can_bus_mlp to encode ego_pose token and send it to LLM BEV: temporal modeling for streampetr.
How to add BEV to the model? According to the paper seem added MLP, but the BEV is so long?
Great work!!!
How to evaluate the results of the open-loop planning?