Full vision pipeline for multiple objects!

[nkumar-bdai]

This PR enables us to run our pipeline with perception directly from vision without making the hard assumption that there is only one object in the world (i.e, the brush). We can now easily modify the numbers and types of objects!

Example Command:

python predicators/main.py --env spot_bike_env --approach spot_wrapper[oracle] --debug --strips_learner oracle --sampler_learner oracle --bilevel_plan_without_sim True --num_test_tasks 1 --explorer active_sampler --spot_robot_ip 10.17.30.30 --spot_grasp_use_sam True --perceiver spot_bike_env --execution_monitor expected_atoms --horizon 15 --spot_initialize_surfaces_to_default False --spot_visualize_vision_model_outputs False --seed 123 --test_task_json_dir predicators/envs/assets/task_jsons/spot_bike_env/brush_screwdriver_perception/

[tsilver-bdai]

halfway through, looking good so far!

[tsilver-bdai]

great work!