Flow-matching robot policies generate each action chunk through repeated network evaluations, one per denoising step. Cross-cycle caches such as C3ache reuse computation from earlier replanning calls, but whether they beat simply taking fewer steps is unclear. We compare the two in Fast-WAM, a 6B-parameter world action model, on LIBERO-Goal. Two uncached steps keep the success of ten: on all 500 initial states they are non-inferior within two percentage points, for one checkpoint and noise seed. Under PyTorch's default compile mode, they also cut a replanning call on an RTX A6000 from 202 to 92 ms. This leaves little room for caching. A cache that keeps two fresh steps cannot be faster than two uncached steps: every call still encodes the images, runs the video expert and computes those two steps. Caching the first of two steps, refreshed every fourth call, saves at most 11--14% of a call by our cost model. Cached policies are also fragile. Their losses tracked the share of the flow path the cached steps cover: under Fast-WAM's current time shift, C3ache's eight cached steps cover 80% of the path instead of 44%, and the cached policy succeeds on 63% of states against 95% uncached. Resampling the noise on every call cost caches covering 60% or more of the path 14--40 points. These results argue for fewer-step baselines and for testing reuse under the deployed sampler and noise settings. We release our evaluation harness, logs and a checklist.
Figures & tables
Figure 1: Success against mean latency per call on 100 LIBERO-Goal states (one RTX A6000). Blue: uncached, 1–10 steps. Orange: caches refreshed every τ calls (squares: 8 of 10 steps cached; diamond: attenuated; triangle: 2 of 4 cached). Bars: 95% Wilson intervals.
path cached (%)
success (%)
mean latency (ms)
p95 latency (ms)
Configuration
K
IDM
orig.
IDM
orig.
IDM
orig.
IDM
orig.
Uncached
1
–
–
88
93
81
81
82
82
Uncached
2
–
–
95
96
94
92
96
95
Uncached
4
–
–
95
95
119
118
125
122
Uncached
10
–
–
95
95
202
202
209
209
Cache, τ=4
4
50
17
96
95
104
105
123
124
Table 1: Success and latency per call on 100 LIBERO-Goal states (one RTX A6000), Optional-IDM (IDM) at s=1 and the original at s=5 , under PyTorch’s default compile mode. Cache latencies include each episode’s opening refresh. Means average episodes’ mean call times; p95 pools all calls.
Figure 2: Success against the cached share of the flow path ( τ=8 , 50 states; 0% is uncached). Triangles: s=3 . Crosses: fresh noise per call. Bars: 95% Wilson intervals.
Appendix figures & tables9 assets
Supplementary material from the paper’s appendix.
Appendix
Ckpt.
Contrast (A vs. B)
A (%)
B (%)
A − B
Tango 90%
p
Holm p
IDM
2 vs. 10 steps
95
95
0
[−4.1,4.1]
1.00
1.00
IDM
2 steps vs. cache τ=8
95
63
+32
[23.3,40.8]
<10−4
<10−4
IDM
2 steps vs. attenuated τ=8
95
81
+14
[7.2,21.6]
0.0026
0.023
IDM
2 steps vs. cache τ=2
95
97
-2
[−6.9,2.5]
0.69
1.00
IDM
2 steps vs. 4-step cache τ=4
95
96
-1
[−4.4,1.7]
1.00
1.00
IDM
attenuated vs. plain τ=8
81
63
+18
[9.1,27.0]
0.0021
0.021
Appendix
Table A1: The eleven primary contrasts of Sec. 3 (100 LIBERO-Goal states). A and B are success rates, the difference A − B is paired by state and given in points, p is an exact McNemar test, and Holm p is adjusted over the family.
Ckpt.
s
cached
share (%)
σ fresh
fixed (%)
fresh (%)
p
IDM
1
0
0
–
96
96
1.00
IDM
1
3
30
0.70
94
–
–
IDM
1
4
40
0.60
94
96
1.00
IDM
1
5
50
0.50
98
–
–
IDM
1
6
60
0.40
96
82
0.039
IDM
1
7
70
0.30
88
62
0.0023
Appendix
Table A2: Every point of Fig. 2 ( τ=8 , 50 LIBERO-Goal states). Cached : leading steps of ten served from the cache. σ fresh : noise level at which fresh computation resumes. Fixed and fresh : success (%) under Fast-WAM’s fixed initial noise and with noise resampled on every call. p : exact McNemar test of fresh against fixed noise, uncorrected.
Condition
Ckpt.
states
10 steps
2 steps
2 − 10, Tango 90%
1 step
cache τ=8
LIBERO-Goal, all states
IDM
500
96.4
96.6
[−1.4,1.8]
–
–
Camera jitter 1.5 cm
IDM
50
66
66
[−9.2,9.2]
68
24
Camera jitter 1.5 cm
IDM
250
64.8
70.0
[1.4,9.3]
–
–
Camera jitter 1.5 cm
orig.
50
48
44
[−14.4,5.9]
–
–
Fresh noise per call
IDM
50
96
96
[−8.1,8.1]
–
26
Fresh noise per call
orig.
50
96
92
[−11.4,1.3]
–
94
Appendix
Table A3: Two against ten uncached steps beyond Table 1 , with one step and the τ=8 cache where run. Success is in percent, and the interval is a Tango 90% interval on the paired difference, in points. The fresh-noise cache column is the default cache at each checkpoint’s own shift. The 250-state jitter row ran on the idle machine and repeats every two- and ten-step outcome of the 50-state row on the states the two share.
Host and checkpoint
1 step
2 steps
4 steps
10 steps
cache τ=8
F/a
Table 1 host, IDM
81
94
119
202
116
4.9
Table 1 host, orig.
81
92
118
202
120
4.8
Table 1 host, orig., CUDA graphs
79
86
100
143
–
10.0
Follow-up host, IDM (round 1)
186
192
–
477
240
3.4
Shared host, IDM
173
185
–
–
–
–
Shared host, orig.
158
190
–
–
–
–
Appendix
Table A4: Mean latency per call (ms) and fixed cost F in units of the per-step cost a . On the Table 1 machine F/a comes from a least-squares fit over 1–10 steps. On the follow-up machine it comes from the 2- and 10-step calls, because its 1-step arm changed load regime halfway through. The idle machine’s row averages its three interleaved rounds.
Host
success (%)
reuse (ms)
refresh (ms)
reuse/refresh
saving (%)
vs. 2 steps
quiet A6000, IDM
94
83
95
0.870
10
–
quiet A6000, orig.
95
83
95
0.869
10
–
shared A6000, IDM
94
165
197
0.838
12
0.93
shared A6000, orig.
95
163
196
0.835
12
0.91
follow-up A6000, IDM, round 1
–
178
211
0.843
12
0.98
follow-up A6000, IDM, round 2
–
154
184
0.835
12
0.80
Appendix
Table A5: The composed cache ( K=2 , step 0 cached, τ=4 ). Success : on the 100 LIBERO-Goal states, where two uncached steps succeed on 95% (IDM) and 96% (orig.). The follow-up and idle rounds timed three states per task. Reuse and refresh : mean call time within the same run. Saving : mean saving per call at τ=4 implied by the two. vs. 2 steps : composed-cache latency over that of two uncached steps timed separately on the same machine. The quiet machine had no such baseline.
Ckpt.
s
noise
cached
τ
share (%)
D
success (%)
pred. (%)
IDM
1
fixed
3
8
30
0.017
94
–
IDM
1
fixed
4
8
40
0.028
94
–
IDM
1
fixed
5
8
50
0.052
98
–
IDM
1
fixed
6
8
60
0.190
96
–
IDM
1
fixed
7
8
70
0.344
88
–
IDM
1
fixed
8
8
80
0.401
54
–
Appendix
Table A6: The cache screen (Sec. 4 ) on the 50 states of Fig. 2 . Cached : leading steps of ten served from the cache, refreshed every τ calls ( τ=0 : only when an episode starts). D : mean RMS difference between the cached and the uncached action chunk over the ten executed actions, in normalized units. Success : the configuration in closed loop on the same states. Below the second rule, configurations never run in closed loop before, on the other 50 states (initial states 5–9 of every task), with the success predicted from D by the logistic fit to the rows above the first rule.
Figure A1: Closed-loop success against the screen’s deviation D for the 24 τ=8 configurations of Table A6 (markers as in Fig. 2 ; diamonds: s=2 ). Grey: the logistic fit on logD that made the held-out predictions. Stars: held-out configurations on the other 50 states. Bars: 95% Wilson intervals.
A
B
A (%)
B (%)
A − B
Tango 90%
p
IDM, 6 of 10 cached, episode latent
fresh noise
96
82
+14
[3.5,25.4]
0.065
IDM, 6 of 10 cached, episode latent
fixed latent
96
96
0
[−8.1,8.1]
1.00
IDM, 8 of 10 cached, episode latent
fresh noise
64
26
+38
[23.3,51.2]
0.0002
IDM, 8 of 10 cached, episode latent
fixed latent
64
54
+10
[0.1,20.8]
0.18
IDM, 2 steps, delay 2
10 steps, delay 5
96
86
+10
[1.6,20.0]
0.12
IDM, 2 steps, delay 2
2 steps, no delay
96
94
+2
[−5.1,9.7]
1.00
Appendix
Table A7: Per-episode latent and delayed execution (50 LIBERO-Goal states; Optional-IDM at s=1 , the original checkpoint at s=5 ). A and B are success rates, the difference A − B is paired by state and given in points with its Tango 90% interval, and p is an exact McNemar test, uncorrected. Episode latent : each episode’s first fresh-noise latent kept for all of its calls. Delay d : the first d of the ten actions executed after a call come from the previous chunk; 5/2 is 5 after refresh calls and 2 after reuse calls. The last block uses the delays of CUDA-graph execution, 2 for two steps and 3 for ten.
IDM
orig.
Task
2 steps
10 steps
τ=8
2 steps
10 steps
τ=8
0
open the middle drawer of the cabinet
10
10
9
10
10
10
1
put the bowl on the stove
9
10
8
10
10
10
2
put the wine bottle on top of the cabinet
9
7
10
10
9
8
3
open the top drawer and put the bowl inside
8
9
3
9
9
9
4
put the bowl on top of the cabinet
10
10
9
10
10
10
Appendix
Table A8: Successes of 10 per LIBERO-Goal task behind Table 1 (one RTX A6000): two and ten uncached steps and the τ=8 cache, Optional-IDM (IDM) at s=1 and the original checkpoint at s=5 .