πR²: Rebuttal Materials

Anonymous submission · under double-blind review · ICLR 2027

R1 · reviewers G5jF, HPaC

Kinetix: dynamic tasks under VLA-level delay

The 6 Kinetix levels whose state includes objects beyond proprioception, which we treat as the slow channel. H = 8, 2,048 rollouts per level, 3 seeds.

Methodd₀ = 1d₀ = 2d₀ = 3d₀ = 4
Standard Flow (async)0.655 ± 0.0000.569 ± 0.001––
RTC0.838 ± 0.0000.690 ± 0.003––
Train-time RTC0.852 ± 0.0170.794 ± 0.006––
FASTER0.826 ± 0.0730.763 ± 0.041––
SDP0.180 ± 0.0100.153 ± 0.0240.088 ± 0.0030.091 ± 0.001
πR² (ours)0.819 ± 0.0040.804 ± 0.0060.769 ± 0.0080.769 ± 0.008
πR² w/o async slow channel0.822 ± 0.0040.807 ± 0.0100.744 ± 0.0290.646 ± 0.017
πR² w/o staircase (= RTC + async VLM)0.812 ± 0.0170.776 ± 0.0210.707 ± 0.0270.642 ± 0.059
πR² w/o latency-adaptive0.210 ± 0.0310.190 ± 0.0350.197 ± 0.0300.197 ± 0.030

Each method is charged its own deployment delay (R4); “–” means that delay no longer fits the chunk (H/2 = 4). πR² and its ablations receive the measured cache age dvis through the delay embedding, as on the robot; the baselines have no cache. Bold: methods at or above 0.80, i.e. comparable at that delay; where none reaches it, the best.

R2 · reviewers HPaC, eSo2

Both conditioning delays vary, and πR² handles it

Left: overhead camera. Right: d, the policy call interval, and dvis, the age of the cached VLM embedding — both re-measured every call. Dashed line: the 5-tick training range.

·

dvis varies continuously around a median of ~2.5 ticks; in five of the twelve rollouts it leaves the 5-tick training range, peaking at 13.9 ticks (2.8× the range) in one. The call interval holds within ±10% on 93–100% of calls and spikes to at most 7.0 ticks when a chunk arrives late. All twelve rollouts succeed.

R3 · reviewers G5jF, eSo2

Measured latency and staleness on the robot

Measured on the robot’s compute, 1 tick = 40 ms. πR² on one GPU still beats every RTC configuration, including RTC with an asynchronous VLM on two.

MethodGPUsVLMPolicy latency d VLM latency dvlmCache age dvis Proprio staleness sproVLM staleness svis svis > 5 ticks
πR²2async31.5 ms (0.79)60.8 ms (1.52)94 ms (2.36)48 ms (1.21)141 ms (3.52)0.0%
πR²1async69.3 ms (1.73)69.4 ms (1.73)88 ms (2.21)104 ms (2.61)194 ms (4.85)40.5%
Train-time RTC1sync133.3 ms (3.33)––203 ms (5.08)203 ms (5.08)50.8%
Train-time RTC2async94.4 ms (2.36)59.8 ms (1.49)96 ms (2.40)142 ms (3.55)235 ms (5.87)83.8%

Ticks in parentheses. Staleness is the median over executed control ticks and counts each action’s offset in its chunk. “–”: already included in d.

R4 · reviewer eSo2

What delay each method is charged in simulation

In simulation the policy latency and cache age are set by a fixed cost model from GR00T-N1.7 — VLM latency dvlm = 0.75d₀ and denoising-step latency ddn = 0.25d₀, where the unit delay d₀ is the cost of GR00T’s default 4 denoising steps — not by how many denoising steps the simulation network runs. Flow and RTC run 15 steps on Leap Cube and up to 5 on Kinetix for quality, but are charged only GR00T’s 4, which favors them. A method is excluded at a given d₀ when its charged policy latency exceeds H/2 (H = 16 on Leap Cube, H = 8 on Kinetix), as it then no longer allows continuous real-time execution.

Leap Cube · H = 16 · d₀ ∈ {1, 2, 3, 5}

MethodCharged policy latency dddvis
Standard Flow, RTC, Train-time RTCdvlm + 4ddn = ⌈1.75d₀⌉2 / 4 / 6 / ––
FASTER⌈0.75d₀ + (n(d)/K)·d₀⌉1 / 3 / 4 / ––
SDP⌈d₀⌉, stalls1 / 2 / 3 / 5–
πR²1·ddn = ⌈0.25d₀⌉1 / 1 / 1 / 21 / 2 / 3 / 4
πR² w/o asyncdvlm + 1·ddn = ⌈1.0d₀⌉1 / 2 / 3 / 5–
πR² w/o staircase4·ddn = ⌈1.0d₀⌉1 / 2 / 3 / 51 / 2 / 3 / 4
πR² w/o latency-adaptive⌈0.25d₀⌉, stalls1 / 1 / 1 / 21 / 2 / 3 / 4

Kinetix · H = 8 · d₀ ∈ {1, 2, 3, 4}

MethodCharged policy latency dddvis
Standard Flow, RTC, Train-time RTCdvlm + 4ddn = ⌈1.75d₀⌉2 / 4 / – / ––
FASTER⌈0.75d₀ + (n(d)/K)·d₀⌉1 / 4 / – / ––
SDP⌈d₀⌉, stalls1 / 2 / 3 / 4–
πR²1·ddn = ⌈0.25d₀⌉1 / 1 / 1 / 11 / 2 / 3 / 3
πR² w/o asyncdvlm + 1·ddn = ⌈1.0d₀⌉1 / 2 / 3 / 4–
πR² w/o staircase4·ddn = ⌈1.0d₀⌉1 / 2 / 3 / 41 / 2 / 3 / 3
πR² w/o latency-adaptive⌈0.25d₀⌉, stalls1 / 1 / 1 / 11 / 2 / 3 / 3

All values in control ticks. “–” in d means the charged latency exceeds H/2, so the schedule cannot fit a clean prefix and a noise tail of length d inside the chunk; in dvis it means the VLM latency is already inside d. For FASTER, n(d) is the ready-step of the last position it emits under its own schedule (α = 0.6, u₀ = 0.9), with K = 15 steps on Leap Cube and K = 5 on Kinetix; because the schedule depends on the delay, d is the smallest self-consistent solution. πR²’s cache age is charged as dvis = ⌈0.75d₀⌉, the lower end of its real-world range.

← Back to project page