refactor!: don't call get_good_view_with_patch_refinement after successful jump (thousandbrainsproject#747)

Author: scottcanoe

benchmarks/logos_on_objects.csv

@@ -1,6 +1,6 @@
 Experiment,Correct Child or Parent (%)|align right,Used MLH (%)|align right,Num Match Steps|align right,Rotation Error (degrees)|align right,Avg Prediction Error|align right,Run Time (mins)|align right
-infer_comp_lvl1_with_monolithic_models,66.67,54.76,316,82.99,0.38,27
-infer_comp_lvl1_with_comp_models,86.90,41.67,41,48.61,0.33,3
-infer_comp_lvl2_with_comp_models,79.05,43.33,36,44.19,0.34,11
-infer_comp_lvl3_with_comp_models,53.43,57.43,37,35.35,0.34,18
-infer_comp_lvl4_with_comp_models,56.32,57.42,35,30.60,0.33,18
+infer_comp_lvl1_with_monolithic_models,60.71,53.57,321,80.71,0.38,28
+infer_comp_lvl1_with_comp_models,84.52,41.67,44,40.76,0.33,3
+infer_comp_lvl2_with_comp_models,79.05,40.48,37,44.91,0.34,9
+infer_comp_lvl3_with_comp_models,56.57,56.57,40,41.50,0.34,19
+infer_comp_lvl4_with_comp_models,58.24,56.59,33,42.52,0.33,18

benchmarks/ycb_10objs.csv

@@ -1,13 +1,13 @@
 Experiment,Correct (%)|align right,Used MLH (%)|align right,Num Match Steps|align right,Rotation Error (degrees)|align right,Run Time (mins)|align right,Episode Run Time (s)|align right
-base_config_10distinctobj_dist_agent,99.29,1.43,41,10.82,2,9
+base_config_10distinctobj_dist_agent,100.00,1.43,39,11.92,2,10
 base_config_10distinctobj_surf_agent,100.00,0.00,27,15.57,3,12
-randrot_noise_10distinctobj_dist_agent,100.00,1.00,38,19.95,3,16
-randrot_noise_10distinctobj_dist_on_distm,99.00,0.00,30,15.74,2,17
-randrot_noise_10distinctobj_surf_agent,98.00,1.00,28,19.99,4,25
-randrot_10distinctobj_surf_agent,100.00,0.00,28,15.06,2,12
-randrot_noise_10distinctobj_5lms_dist_agent,100.00,0.00,51,48.04,6,45
+randrot_noise_10distinctobj_dist_agent,99.00,4.00,44,21.57,3,16
+randrot_noise_10distinctobj_dist_on_distm,98.00,5.00,30,13.53,3,12
+randrot_noise_10distinctobj_surf_agent,98.00,1.00,28,19.99,4,24
+randrot_10distinctobj_surf_agent,100.00,0.00,28,15.06,2,13
+randrot_noise_10distinctobj_5lms_dist_agent,99.00,0.00,52,54.13,5,41
 base_10simobj_surf_agent,93.57,10.71,82,13.93,6,31
-randrot_noise_10simobj_dist_agent,81.00,34.00,204,28.86,9,76
-randrot_noise_10simobj_surf_agent,83.00,33.00,166,31.98,18,145
+randrot_noise_10simobj_dist_agent,81.00,38.00,211,33.62,12,96
+randrot_noise_10simobj_surf_agent,83.00,33.00,166,31.98,17,139
 randomrot_rawnoise_10distinctobj_surf_agent,71.00,74.00,14,96.27,4,6
-base_10multi_distinctobj_dist_agent,81.43,17.86,32,18.06,45,2
+base_10multi_distinctobj_dist_agent,56.43,2.14,33,26.53,33,1

benchmarks/ycb_77objs.csv

@@ -1,6 +1,6 @@
 Experiment,Correct (%)|align right,Used MLH (%)|align right,Num Match Steps|align right,Rotation Error (degrees)|align right,Run Time (mins)|align right,Episode Run Time (s)|align right
-base_77obj_dist_agent,93.51,10.82,98,13.59,21,62
-base_77obj_surf_agent,99.57,7.36,57,8.75,15,38
-randrot_noise_77obj_dist_agent,86.58,24.68,162,30.62,35,116
-randrot_noise_77obj_surf_agent,90.48,23.81,111,31.05,35,115
-randrot_noise_77obj_5lms_dist_agent,92.21,1,72,53.19,12,119
+base_77obj_dist_agent,93.07,15.58,109,16.93,20,61
+base_77obj_surf_agent,99.57,7.36,57,8.75,14,36
+randrot_noise_77obj_dist_agent,87.01,28.14,158,32.66,35,108
+randrot_noise_77obj_surf_agent,90.48,23.81,111,31.05,33,106
+randrot_noise_77obj_5lms_dist_agent,92.21,1,66,57.63,11,106

benchmarks/ycb_unsupervised_inference.csv

@@ -1,3 +1,3 @@
 Experiment,Correct (%)|align right,Num Match Steps|align right,Run Time (mins)|align right,Episode Run Time (s)|align right
-unsupervised_inference_distinctobj_dist_agent,99.00,100,11,7
-unsupervised_inference_distinctobj_surf_agent,100.00,99,19,11
+unsupervised_inference_distinctobj_dist_agent,99.00,96,8,5
+unsupervised_inference_distinctobj_surf_agent,100.00,99,17,10

docs/how-monty-works/policy.md

@@ -64,7 +64,7 @@ As the positioning procedures were implemented in the early development of Monty
 1) Positioning procedures should only be called in `pre_episode`, i.e. before the episode begins.
 2) Positioning procedures may make use of privileged information such as the view-finder and semantic-sensor, which are not available to the learning module. However, consistent with point (1), these should not be leveraged during learning or inference by the Monty agent.
 
-These requirements are currently not enforced in the use of `GetGoodView` and `touch_object` when an agent jumps to a location using a model-based policy. Similarly, `touch_object` is used by the surface agent if it loses contact with the object and cannot find it. Appropriately separating out the role of positioning procedures via the above requirements will clarify their role, and enable policies that do not make use of privileged information, but which serve similar purposes during an experiment.
+These requirements are currently not enforced in the use of `touch_object` when an agent jumps to a location using a model-based policy. Similarly, `touch_object` is used by the surface agent if it loses contact with the object and cannot find it. Appropriately separating out the role of positioning procedures via the above requirements will clarify their role, and enable policies that do not make use of privileged information, but which serve similar purposes during an experiment.
 
 
 # Input-Driven and Hypothesis-Driven Policies

src/tbp/monty/frameworks/environments/embodied_data.py

@@ -610,9 +610,6 @@ def get_good_view_with_patch_refinement(self) -> bool:
         "patch" or "patch_0") to ensure that the patch's central pixel is on-object.
         Up to 3 reorientation attempts are performed using the central patch.
 
-        Also currently used by the distant agent after a "jump" has been initialized
-        by a model-based policy.
-
         Returns:
             Whether the sensor is on the object.
 
@@ -754,9 +751,6 @@ def handle_successful_jump(self):
             self.motor_system._policy.action_details["avoidance_heading"].append(False)
             self.motor_system._policy.action_details["z_defined_pc"].append(None)
 
-        else:
-            self.get_good_view_with_patch_refinement()
-
     def handle_failed_jump(self, pre_jump_state, first_sensor):
         """Deal with the results of a failed hypothesis-testing jump.