← 研踪 PaperTrailarXiv:2611.26385 · v1
Beyond the Sink Interpretation: Radial Attention in Flow-Based VLA Policies
首次提交 2026-11-20 00:41
版本更新 2026-11-20 00:41
雷达观测 2026-11-20 12:52
摘要
Prior work calls the high attention on the first and last action tokens of a flow-based VLA policy a radial action sink, and uses these tokens to set the execution horizon. In language models, a sink is a token with high incoming attention, a small value, and little effect on the output. We test whether that profile holds, not the horizon rule, in one checkpoint on five RoboCasa tasks. The two endpoints receive 5.4% of action-key mass, 1.4 percentage points above their uniform share, and rotary positional encoding produces this excess: directly in shallow layers and through upstream action-token RoPE deeper in the network. The endpoints do not fit the sink profile. Their value norms stay near the action-token mean in all 18 layers. On 250 paired episodes, masking them or zeroing their values changes success by less than 4 points, and replacing their value directions with those of another observation lowers it by 3.2 to 7.2 points across three donors, whereas the same replacement on two non-endpoint tokens in the executed prefix lowers it by 28.4 points. The token that does fit the sink profile on mass and value norm is the prefix BOS; on this checkpoint, zeroing its value raises success in two paired runs, by +13.2 points (95% CI [+7.0, +19.4]) and +6.0 points ([+0.2, +11.8]). Inside the action chunk, a relative-offset routing profile estimated on other tasks keeps 75% of action-token spread and keeps closed-loop success on three held-out tasks (42.7% against a 38.7% baseline), while uniform or shuffled routing drops success to 3-4%. High attention mass alone does not tell what a token does.
作者备注:17 pages
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-20 00:41 · 最新观测版本
Beyond the Sink Interpretation: Radial Attention in Flow-Based VLA Policies
Prior work calls the high attention on the first and last action tokens of a flow-based VLA policy a radial action sink, and uses these tokens to set the execution horizon. In language models, a sink is a token with high incoming attention, a small value, and little effect on the output. We test whether that profile holds, not the horizon rule, in one checkpoint on five RoboCasa tasks. The two endpoints receive 5.4% of action-key mass, 1.4 percentage points above their uniform share, and rotary positional encoding produces this excess: directly in shallow layers and through upstream action-token RoPE deeper in the network. The endpoints do not fit the sink profile. Their value norms stay near the action-token mean in all 18 layers. On 250 paired episodes, masking them or zeroing their values changes success by less than 4 points, and replacing their value directions with those of another observation lowers it by 3.2 to 7.2 points across three donors, whereas the same replacement on two non-endpoint tokens in the executed prefix lowers it by 28.4 points. The token that does fit the sink profile on mass and value norm is the prefix BOS; on this checkpoint, zeroing its value raises success in two paired runs, by +13.2 points (95% CI [+7.0, +19.4]) and +6.0 points ([+0.2, +11.8]). Inside the action chunk, a relative-offset routing profile estimated on other tasks keeps 75% of action-token spread and keeps closed-loop success on three held-out tasks (42.7% against a 38.7% baseline), while uniform or shuffled routing drops success to 3-4%. High attention mass alone does not tell what a token does.
查看此版本解读 →查看此版本原文 ↗