ActionGround: Training-Free Runtime Refinement of Frozen VLA Policies

Namai Chandra; Madhur Thareja; Shriram Damodaran; Addison Lin Wang

cs.ROSep 27, 2026

ActionGround: Training-Free Runtime Refinement of Frozen VLA Policies

Authors: Namai Chandra, Madhur Thareja, Shriram Damodaran, Addison Lin Wang

Organizations: Indian Institute of Technology Madras · Indian AI Research Organisation · Nanyang Technological University

Abstract

Vision-Language-Action (VLA) models map visual observations and language instructions directly to robot actions, but they do not explicitly represent the phase structure of manipulation tasks or the rigid-body dynamics governing execution. We present ActionGround, a neuro-symbolic, training-free runtime layer that wraps a frozen VLA policy without retraining, fine-tuning, or weight access, adding less than 1 ms of overhead per control step. A symbolic phase-aware finite-state machine identifies the manipulation phase (approach, grasp, transport, or place) and applies a phase-specific rule-based correction. In parallel, an always-on, inertia-weighted Euler-Lagrange term incorporates the robot's equations of motion into each control step, while its dynamics residual is logged as a consistency diagnostic rather than used as a gate. We evaluate ActionGround across OpenVLA, OpenVLA-OFT, Force-VLA, and Generalist-VLA on ten LIBERO-Spatial pick-and-place tasks using a 7-DoF Franka Panda. With fixed parameters across tasks and backbones, ActionGround improves success rate by up to 6 percentage points and stability by up to 19.3 percentage points, while improving trajectory efficiency by up to 15%. In a separate Robosuite noise sweep, ActionGround provides approximately a 10x improvement in trajectory-jerk robustness under injected action noise. In a matched-seed Robosuite simulation companion to a real Agilex Piper trial, simulated baseline success increases from 35% to 95%. The physical-hardware experiment is presented as a qualitative deployment demonstration; quantitative per-trial success on the real arm is left for future work. Our evaluation is limited to rigid-object pick-and-place manipulation.

Figures & tables

Fig. 1 : OpenVLA backbone, same weights, on the matched-seed Robosuite place-on-plate simulation companion described in Sec. IV-B ( $n{=}20$ per mode). ActionGround succeeds where the unmodified backbone fails under the injected placement-phase bias.

Fig. 2 : Overview of our ActionGround framework . See main texts for details.

Strategy	Representative works	Guarantee	Runtime
Soft Residual	PINNs [ 10 ] , PIPER [ 51 ]	Approximate	No
Hard Structural	DeLaN [ 11 ] , HNNs [ 13 ] , SymODEN [ 14 ]	Exact	No
Stability-Certified	Neural Lyapunov [ 45 ] , CBF [ 15 ]	Certified	Optional
ActionGround (ours, runtime)	FSM + EL residual (this work)	Diagnostic	Yes

TABLE I : Taxonomy of physics-constraint strategies in robotic policy learning. “Runtime” indicates whether the constraint is active at inference.

Backbone	$\Delta$ Stab	$\Delta$ Succ
OpenVLA	$+16.7\%$	$+2\%$
Force-VLA	$+18.2\%$	$+4\%$
Generalist-VLA	$+19.3\%$	$+6\%$
OpenVLA-OFT †	$+2.8\%$	$+0\%$

TABLE II : ActionGround gain over Baseline on LIBERO-Spatial (aggregate, %).

Fig. 3 : ActionGround’s runtime refinement layer. Channel A (symbolic phase grounding) and Channel B (always-on, inertia-weighted Euler–Lagrange correction) each propose a candidate correction; the two are combined and capped-blended with the frozen VLA’s own action. The Euler–Lagrange residual $r_{\text{EL}}$ (dashed branch) is a logged diagnostic only and does not gate or scale the executed correction. See Sec. III-C for details.

Backbone	Base		Temp		ActionGround
Backbone	Stab %	Succ %	Stab %	Succ %	Stab %	Succ %
OpenVLA (single-step)	20.1	36	36.2	28	36.8	38
OpenVLA-OFT (chunked)	86.1	92	88.1	92	88.9	92
Force-VLA (force-residual)	20.0	40	36.6	36	38.2	44
Generalist-VLA (flow-matching)	29.9	36	39.2	26	49.2	42

TABLE III : Aggregate stability and success rate on LIBERO-Spatial (unweighted mean over T0–T9,

5

trials per task at seed 7) under three inference-time modes. Our ActionGround achieves aggregate improvements over the same-row Baseline.

Fig. 4 : Cross-architecture qualitative comparison on LIBERO-Spatial. Baseline (top) vs. ActionGround (bottom), Force-VLA on T6/T8/T2/T9 (cols 1–4) and Generalist-VLA on T3/T8/T6/T4 (cols 5–8). Bottom-right badge: per-task success ( $n{=}5$ ). Top-right green badge: strict per-task recovery under ActionGround.

Fig. 5 : Real-world Agilex Piper pick-and-place under Baseline (top row) and ActionGround (bottom row), shown from a side view (left) and a top view (right).

Explore similar work

PhysVLA: Towards Physically-Grounded VLA for Embodied Robotic Manipulation
Jun 11, 2026Namai Chandra, Shriram Damodaran, Lin WangRobotic Control Robotic Manipulation
Electronic Systems IIT Madras, India · EmPACT Lab Nanyang Technological University Singapore
PHR-VLA: Planning Horizon Reasoning for Vision-Language-Action Models
Aug 27, 2026Davood Soleymanzadeh, Kaidi Zhang, Zhiyuan Zhang +4Vision-Language-Action Models Reasoning for VLA Models
J. Mike Walker ’66 Department of Mechanical Engineering, Texas A&M University, College Station, TX 77843, USA · Department of Industrial Engineering, Purdue University, West Lafayette, IN 47907, USA · Zachry Department of Civil and Environmental Engineering, Texas A&M University, College Station, TX 77843 USA
Harness VLA: Steering Frozen VLAs into Reliable Manipulation Primitives via Memory-Guided Agents
Jul 9, 2026Yixian Zhang, Huanming Zhang, Feng Gao +14Long-Horizon Robotic Manipulation Language-Conditioned Robot Manipulation
Tsinghua University · Striding AI · Purdue University +4

Jun 11, 2026J/K move · Enter open · S save

PhysVLA: Towards Physically-Grounded VLA for Embodied Robotic Manipulation

Namai Chandra, Shriram Damodaran, Lin Wang

Electronic Systems IIT Madras, India · EmPACT Lab Nanyang Technological University Singapore

Robotic Control Robotic Manipulation Vision-Language-Action Models

Open paper AskPDF arXiv

Vision-Language-Action (VLA) models excel at mapping visual inputs and natural language instructions directly to robotic control policies. However, because they are trained primarily to fit behavioural demonstration data, they do not explicitly enforce fundamental physical principles such as rigid-body dynamics or contact constraints. This exposes a critical physics gap: standard temporal smoothing applied on top of single-step or chunked VLAs trades trajectory quality for added failures that short-term memory cannot resolve. To bridge this gap, we introduce PhysVLA (Physics-VLA), a plug-and-play, inference-time framework designed to wrap any frozen VLA backbone without retraining, fine-tuning, or weight access, with less than 1 ms of overhead per control step. PhysVLA intercepts the predicted control action, captures only the simulator or system state, and applies a dual-layered correction: (i) a phase-aware finite-state machine that structures discrete task segments (approach, grasp, transport, and place), and (ii) a selective Euler-Lagrange gate that activates only when a dynamics oracle detects kinodynamic inconsistency. Evaluated across OpenVLA, OpenVLA-OFT, Force-VLA, and Generalist-VLA on LIBERO-Spatial with a 7-DoF Franka Panda, the framework delivers absolute success rate increases of up to 17% and stability increases of up to 19% with no per-task regressions, improves trajectory efficiency by up to 15% across all four backbones, and shows up to a 10x improvement in trajectory jerk robustness on a Robosuite Lift cross-simulator sweep. We further validate the framework on a real Agilex Piper arm with a pick-and-place task, confirming that PhysVLA transfers to physical hardware without retraining, with success-rate improvements of up to 50%, establishing physical awareness as a composable, backbone-agnostic runtime module.

Abstract

Explore similar work

PhysVLA: Towards Physically-Grounded VLA for Embodied Robotic Manipulation

PHR-VLA: Planning Horizon Reasoning for Vision-Language-Action Models

Harness VLA: Steering Frozen VLAs into Reliable Manipulation Primitives via Memory-Guided Agents