This is a very inspiring work.
I noticed there is a config file named sparseoccvla_stage1_3d_600q_lidar_navsim.py, but it seems the corresponding results are not included in the preprint. May I ask whether this configuration uses LiDAR supervision to train OPUS, and then feeds the trained OPUS queries into the LLM for the second-stage end-to-end training? I would also be curious to know the performance achieved with this setup.
This is a very inspiring work.
I noticed there is a config file named
sparseoccvla_stage1_3d_600q_lidar_navsim.py, but it seems the corresponding results are not included in the preprint. May I ask whether this configuration uses LiDAR supervision to train OPUS, and then feeds the trained OPUS queries into the LLM for the second-stage end-to-end training? I would also be curious to know the performance achieved with this setup.