sort_toy_by_weight: .h5 files raw-only (drop the derived *_hand_tactile_force_zeroed and the tare attrs; zero with xhand_dataprocessing/zero_tactile.py)
plug_into_power_strip: .h5 files raw-only (drop the derived *_hand_tactile_force_zeroed and the tare attrs; zero with xhand_dataprocessing/zero_tactile.py)
pick_cup_full_empty: .h5 files raw-only (drop the derived *_hand_tactile_force_zeroed and the tare attrs; zero with xhand_dataprocessing/zero_tactile.py)
pick_bag_place_box: .h5 files raw-only (drop the derived *_hand_tactile_force_zeroed and the tare attrs; zero with xhand_dataprocessing/zero_tactile.py)
held_out_set/toy_shelf: .h5 files raw-only (drop the derived *_hand_tactile_force_zeroed and the tare attrs; zero with xhand_dataprocessing/zero_tactile.py)
held_out_set/tissue: .h5 files raw-only (drop the derived *_hand_tactile_force_zeroed and the tare attrs; zero with xhand_dataprocessing/zero_tactile.py)
held_out_set/sort_toy_by_weight: .h5 files raw-only (drop the derived *_hand_tactile_force_zeroed and the tare attrs; zero with xhand_dataprocessing/zero_tactile.py)
held_out_set/plug_in: .h5 files raw-only (drop the derived *_hand_tactile_force_zeroed and the tare attrs; zero with xhand_dataprocessing/zero_tactile.py)
held_out_set/handover: .h5 files raw-only (drop the derived *_hand_tactile_force_zeroed and the tare attrs; zero with xhand_dataprocessing/zero_tactile.py)
held_out_set/bag_box: .h5 files raw-only (drop the derived *_hand_tactile_force_zeroed and the tare attrs; zero with xhand_dataprocessing/zero_tactile.py)
held_out_set/backpack: .h5 files raw-only (drop the derived *_hand_tactile_force_zeroed and the tare attrs; zero with xhand_dataprocessing/zero_tactile.py)
xhand_data docs: held-out counts now 10 per task (24 sort_toy) from held_out_set/, wrist note in every card, cup durations of the released 100, gap threshold 0.25 s, zeroed-tactile and head_depth coverage, glove name, swap_wrists comment, stats.json scope
splits.json: heldout = the held_out_set test episodes (10 per task, 24 for sort_toy_by_weight); earlier held-out episodes listed as unused; README and cup card to match
GMP_NOWALLS=1 ./run.sh: no maze-wall blocks; a steering bound keeps every planned position within 0.2 maze units of the diffusion estimate instead (globmp_mpc.py: GMP_TUBE; widened where the estimate lies inside a keep-out, doubled when a plan's first solve is infeasible). 10 seeds x 10 episodes per map: SR 1.000 on both maps, score 1.613 (broad) and 1.617 (narrow), every plan clear of the keep-outs at every t; 39 and 19 of 100 plans cross a wall. results/globmp_nowalls holds the 200 episodes; check.py summarizes both folders. The default run is unchanged.
Maze2D (GlobMP in DiRecT's sampler): shorter time limit for gradient projection, results re-run
run.sh: gradient projection gets 0.125 s in a plan's first solve and 0.0625 s in the later ones (GMP_ALG2COLD, new GMP_ALG2WARM; was 1 s and 0.5 s). On 10 seeds x 10 episodes per map the plans hold the same margins, safety, score and Gen. columns are unchanged, and the planning time is 5.7 s (broad) and 7.4 s (narrow) instead of 7.3 s and 9.3 s. results/ holds the new 200 episodes.
9 days ago
Docs: algorithms named, not numbered (paper: INIT Algorithm 1, repair of Eq. (5) Algorithm 2, gradient projection Algorithm 3)
Comments, docstrings, help strings, printed labels and READMEs now say INIT, the repair of Eq. (5) and gradient
projection instead of algorithm numbers, so the text does not depend on the paper's numbering. The root README
keeps one line with the paper's numbers (and the cone projection, Algorithm 4).
No code, identifier, option or result key changed. The one stored string that changed is the "algorithm" field
that globmp.synthetic.run writes for new runs; the reference files are unchanged.
Algorithm 2 as written in the paper: step 1/beta and the repair of Eq. (5)
- globmp/repair.py (new): the repair of Algorithm 2 (Algorithm 1 in the paper), Eq. (5). After a cost step Y, the
next curve is the point nearest to Y of the set where every block, linearized at the current feasible curve, is
nonnegative on its whole interval. The solve keeps a list of contacts (block, instant): find the instants at
which a linearized block is negative, add them, fit nonnegative weights of all listed contacts jointly, repeat
until none is negative. Problem.repair(Y, z, tv); Problem.repair_tol.
- globmp/gpa.py: Problem.solve repairs with it (Problem.alg2_repair = 'eq5', the default). 'restore' keeps the
restoration of restore.py as the repair; the scaling study and Maze2D set it, because their results were
obtained with it, and the 'sampled' and 'orthant' modes always use it. The restoration still makes an
infeasible seed feasible.
- Section 4.1 (globmp/synthetic/run.py): one configuration, Algorithm 2 with step 1/beta and this repair from
INIT's curve; …
17 days ago
Add plug_into_power_strip (100 train + 10 held out); README, stats.json, make_stats.py; placing tasks labelled place, insert kept for the plug