SWAP: Stepwise Action Policy Routing for Vision-Language-Action Models

Mousumi Das; Aditeya Prajapati; Abrar Anwar; Jesse Thomason

cs.ROOct 3, 2026

SWAP: Stepwise Action Policy Routing for Vision-Language-Action Models

Authors: Mousumi Das, Aditeya Prajapati, Abrar Anwar, Jesse Thomason

Organizations: University of Southern California

Abstract

Robot manipulation systems using Vision-Language-Action (VLA) model backbones typically use just one VLA for task execution. However, individual VLAs do not perform well across different task states and environments. We introduce a framework for dynamically composing multiple VLA policies during execution: StepWise Action Policy Routing (SWAP). SWAP formulates policy routing as an offline reinforcement learning problem, learning a routing critic that selects the most appropriate policy at each decision step given the current observation. SWAP enables robots to select new policies to execute online rather than committing to a single policy for the duration of an episode. We evaluate SWAP on both real-world DROID manipulation tasks and LIBERO simulation experiments. SWAP improves over fixed-policy execution and routing baselines, giving absolute improvements in real-world task success up to 33% while reducing successful trajectory robot action step length by 28.3%.

Figures & tables

Fig. 1 : Policy routing for robot manipulation. We introduce SWAP, an offline reinforcement learning framework for online policy routing. Given candidate robot policies, SWAP dynamically selects the best policy online for an instruction and current observation during task execution.

Fig. 2 : Overview of SWAP. (a) We construct an offline routing dataset by collecting trajectories from a set of candidate robot policies and chunking them into routing transitions labeled with sparse rewards and termination signals. (b) SWAP trains a routing critic offline using IQL, where visual observations and task instructions are encoded with a frozen vision-language model and combined with a candidate policy embedding to predict policy-specific Q-values. (c) SWAP evaluates all candidate policies under the current observation, selects the highest-valued policy, executes it in the environment for a short horizon, and repeats this process online throughout task execution.

Fig. 3 : Evaluation environments. We evaluate SWAP on both real-world DROID [ 1 ] robot manipulation tasks and LIBERO [ 48 ] simulation tasks. For simulation, we use the spatial subset of LIBERO-Plus [ 18 ] , which includes challenging out-of-distribution task variations.

Task	Succ.	Steps	Succ.	Steps	Succ.	Steps	Succ.	Steps	Succ.	Steps
	Individual Policy						Routing
	$\pi_{0}$		$\pi_{0.5}$		$\pi_{0}$ -FAST		Random		SWAP (Ours)
Marker $\rightarrow$ Bowl	30%	274	60%	249	80%	209	60%	233	100%	200
Cube $\rightarrow$ Cup	50%	358	20%	420	40%	372	60%	336	90%	224
Cube on Cloth $\rightarrow$ Bowl	40%	288	50%	201	70%	227	80%	262	100%	155
Average	40%	307	43.3%	290	63.3%	269	66.7%	277	96.7%	193

TABLE I: Real Robot Policy Routing. We present real-world policy routing results on a Franka DROID setup across three manipulation tasks. SWAP achieves the strongest overall performance, improving success rate from 63.3% for the best fixed policy baseline to 96.7%, demonstrating that learned policy routing can effectively compose complementary behaviors across candidate policies. Additionally, among successful trajectories, SWAP completes tasks in 28.3% fewer steps than the best-performing individual policy (193 vs. 269 steps).

Segment	$\pi_{0}$ -FAST	$\pi_{0.5}$
First Third	51.0%	49.0%
Middle Third	22.6%	77.4%
Final Third	36.3%	63.7%

TABLE II: Policy preference shifts as execution progresses. Policy selection frequencies across normalized trajectory segments. SWAP begins with nearly equal use of both policies but concentrates on

\pi_{0.5}

during later stages of execution. This suggests that SWAP discovers policy strengths and exploits them at different phases of task completion.

	Individual Policy			Routing
Task with Distractors	$\pi_{0}$	$\pi_{0.5}$	$\pi_{0}$ -FAST	SWAP (Ours)
Marker $\rightarrow$ Bowl	10%	40%	40%	90%
Cube on Plate $\rightarrow$ Bowl	10%	30%	0%	80%
Average	10%	35%	20%	85%

TABLE IV : Real Robot Policy Routing under OOD Distractors. We add unseen distractor objects and evaluate each method over 10 trials per task. SWAP maintains high success despite the distribution shift.

Fig. 4 : Q-values and policy routing on the DROID marker-in-bowl task. SWAP scores $\pi_{0}$ -FAST, $\pi_{0}$ , and $\pi_{0.5}$ based on the predicted Q-values of each policy throughout task execution. The images highlight key routing decisions in one SWAP rollout. The plot shows the corresponding Q-values over time, as well as the policy selected by SWAP at each decision step. Initially, SWAP favors $\pi_{0}$ -FAST, which takes larger steps during free-space movement. As the robot approaches the object, it switches to $\pi_{0.5}$ for the more precise grasp. After the marker is accidentally dropped, the Q-values collapse and SWAP again selects $\pi_{0.5}$ to recover and reattempt the grasp. Once the marker is secured, SWAP switches back to $\pi_{0}$ -FAST for faster motion to the goal.

Robotic Manipulation

Robotic manipulation studies how robots perceive, plan, and control physical interactions to grasp, move, assemble, and use objects.

See all 1,671 papers

LayerRoute: Action-Conditioned Mixture-of-Layers Routing for Vision-Language-Action Policies
Sep 5, 2026Zheng Lu, Haoran Liao, Wanqi Zhong +9Visuomotor Policy Learning Robotic Manipulation
Tsinghua University · Zhejiang University · Nanyang Technological University +2
Fast-VLA: You Only Think Once for Robot Manipulation
Oct 15, 2026Kehong Gong, Guanli Hou, Xiaoxu Zheng +6Vision-Language-Action Models Closed-Loop Control of VLA Models
Huawei Technologies Co., Ltd · Huawei Central Research Institute
FocusVLA: Future-Intent Guided Dynamic View Routing for Vision-Language-Action Models
Oct 22, 2026Huixin Xu, Ziya Yu, Peng LiVision-Language-Action Models Vision-Based Robot Control
Nanjing Institute of Software Technology · Institute of Software, Chinese Academy of Sciences · University of Chinese Academy of Sciences, Nanjing +2

Sep 5, 2026J/K move · Enter open · S save

LayerRoute: Action-Conditioned Mixture-of-Layers Routing for Vision-Language-Action Policies

Zheng Lu, Haoran Liao, Wanqi Zhong, Yunhe Ni+8

Tsinghua University · Zhejiang University · Nanyang Technological University+2

Visuomotor Policy Learning Robotic Manipulation Vision-Language-Action Models

Open paper AskPDF arXiv

Vision-Language-Action (VLA) policies leverage pretrained vision-language models (VLMs) to guide action generation for robot control. VLMs provide hierarchical visual-semantic representations that evolve across layers, from local visual geometry to abstract, language-aligned semantics; different manipulation tasks may therefore require different mixtures of layer representations. Meanwhile, the action module maintains intermediate representations that evolve throughout action computation and may provide useful information for subsequent decisions. However, existing VLA interfaces offer limited flexibility in representation access: VLM information is exposed through fixed layer assignments for each action layer, while intermediate action states are only propagated implicitly through residual streams without explicit reuse. We introduce LayerRoute, an action-conditioned representation routing interface that enables adaptive access to VLM layers and action representations. The Layer Mixture Router dynamically forms mixtures of cached VLM representations, while Action-State Reread reuses earlier action representations. Across diverse simulation and real-world benchmarks, LayerRoute consistently improves StarVLA-

π

and

π_{0.5}

, achieving up to 7.2 gains on LIBERO Long with only 0.31% / 3.87% additional parameters. Ablation studies validate the benefit of action-conditioned layer routing, while routing analyses reveal structured allocation patterns across action layers and task settings.

Abstract

Robotic Manipulation

Explore similar work

LayerRoute: Action-Conditioned Mixture-of-Layers Routing for Vision-Language-Action Policies

Fast-VLA: You Only Think Once for Robot Manipulation

FocusVLA: Future-Intent Guided Dynamic View Routing for Vision-Language-Action Models