Under Construction.
Figure 6.1a grab(1)
level=1
Figure 6.1b. grab(8)
Nnboxes=9, level=1
Figure 6.2a place() uniform policy
No constaints, learns P, level=1
Figure 6.2b place()
P from 6.2a, near-greedy, level=1
Figure 6.3a avoid_left(3), level=1
Figure 6.3b place_on(5), level=1
Figure 6.4a grab_covered(3), level=2
No constraints, actions = grab, place
Figure 6.4b grab_covered(3), level=2
No constraints, nboxes=9
Figure 6.5a pick_and_place_covered(3)
With constraints, level=2
Figure 6.5b grab_covered(3), level 2
No constraints, policy from 6.5a
Figure 6.6a level=2, no constraints
pick_and_place_w_obstacles(p)
Near-greedy use of place()
Figure 6.6b level=2, w constriants
Otherwise same as 6.6a
Figure 6.6c Level=2
Greedy grab, place,
plus basic actions
Figure 6.7 Build stack height 5
level=2, grab(q), place_on(r)
Figure 6.8 Stack 1 on 5, level=2
Repurposes avoid_l/r(q), grab(r)
Figure 6.9a grab_shape(p), level=2
add basic actions
Figure 6.9b grab_shape(p), level=2
add basic rotations
Figure 6.10 grab_lift_shape(p)
level=3, uses grab_shape(p) 6.9
Figure 6.11 grab_shape_hr(p))
Figure 6.12a Level=2
pick_from_pile_and_place(p)
Figure 6.12b Level=2
Figure 6.12c Level=2
Figure 6.12d Level=2
Figure 6.13a terrain_follow(p)
Optimized path
Figure 6.13b terrain_follow(p)
Non-optimized path
Figure 6.14a lunar lander
Sub-problem for GA
Figure 6.14b lunar lander
Full solution using GA
Figure B.3 checkerboard
Optimization steps