Device-Conditioned Neural Architecture Search for Efficient Robotic Manipulation

Yiming Wu; Huan Wang; Zhenghao Chen; Ge Yuan; Dong Xu

cs.ROApr 11, 2026

Device-Conditioned Neural Architecture Search for Efficient Robotic Manipulation

Authors: Yiming Wu, Huan Wang, Zhenghao Chen, Ge Yuan, Dong Xu

Organizations: The University of Hong Kong, China · Westlake University, China · The University of Newcastle, Australia

Abstract

The growing complexity of visuomotor policies poses significant challenges for deployment with heterogeneous robotic hardware constraints. However, most existing model-efficient approaches for robotic manipulation are device- and model-specific, lack generalizability, and require time-consuming per-device optimization during the adaptation process. In this work, we propose a unified framework named Device-Conditioned Quantization-For-All (DC-QFA) which amortizes deployment effort with the device-conditioned quantization-aware training and hardware-constrained architecture search. Specifically, we introduce a single supernet that spans a rich design space over network architectures and mixed-precision bit-widths. It is optimized with latency- and memory-aware regularization, guided by per-device lookup tables. With this supernet, for each target platform, we can perform a once-for-all lightweight search to select an optimal subnet without any per-device re-optimization, which enables more generalizable deployment across heterogeneous hardware, and substantially reduces deployment time. To improve long-horizon stability under low precision, we further introduce multi-step on-policy distillation to mitigate error accumulation during closed-loop execution. Extensive experiments on three representative policy backbones, such as DiffusionPolicy-T, MDT-V, and OpenVLA-OFT, demonstrate that our DC-QFA achieves $2\text{-}3\times$ acceleration on edge devices, consumer-grade GPUs, and cloud platforms, with negligible performance drop in task success. Real-world evaluations on an Inovo robot equipped with a force/torque sensor further validates that our low-bit DC-QFA policies maintain stable, contact-rich manipulation even under severe quantization.

Figures & tables

Figure 1: Overview of the proposed DC-QFA pipeline. In stage I (Quantized Supernet Training), we train a weight-sharing supernet with quantization-aware training. In stage II (Device-aware Subnet Search), we search subnets under device-specific latency and memory constraints by optimizing a multi-objective objective guided by measured costs. In stage III (Deployment), we export the selected subnet, calibrate/compile for the target backend, and perform low-bit inference on-device.

Figure 2: Experimental setups in simulation and the real world. (a) Push-T. (b) LIBERO. (c) CALVIN. (d) Real robot tasks, including object handling (egg) and tool use (brush).

Method	Bit-width	Model Size	LIBERO
Method	Bit-width	Model Size	Spatial	Object	Goal	Long	Avg $\uparrow$
OpenVLA-OFT	FP16	15.1 GB	97.6	98.4	97.9	94.5	97.1
w/ BatchQuant	FP8	7.5 GB	96.8	98.2	96.2	93.2	96.1
DC-QFA (Ours)			97.2	98.0	97.2	93.9	96.6
w/ SmoothQuant	W8A8	7.5 GB	94.8	93.1	93.4	90.7	93
w/ AWQ			94.7	93.5	93.8	91.1	93.3

Table 1: Performance of OpenVLA-OFT [ 16 ] on the LIBERO [ 19 ] benchmark. We compare FP16 and quantized variants at FP8/W8A8 and W4A4 using SmoothQuant [ 32 ] , AWQ [ 18 ] , SQIL [ 24 ] , and BatchQuant [ 3 ] . Metrics are success rates (%) on four subsets; Avg is the mean (higher is better). DC-QFA (ours) is highlighted and matches or surpasses SQIL at the same bit-width.

Method	Bit-width	Model Size	Success Rate $\uparrow$
DiffusionPolicy-T	FP16	38.48 MB	0.772 $\pm$ 0.039
w/ BatchQuant	FP8	19.24 MB	0.759 $\pm$ 0.021
DC-QFA (Ours)			0.770 $\pm$ 0.016
w/ SmoothQuant	W8A8	19.24 MB	0.729 $\pm$ 0.011
w/ AWQ			0.733 $\pm$ 0.022
w/ SQIL			0.766 $\pm$ 0.019

Table 2: Results of DiffusionPolicy-T [ 14 ] on the Push-T dataset, where we compare various quantization techniques and their impact on model size and success rate.

Method	Bit-width	Model Size	CALVIN		LIBERO
Method	Bit-width	Model Size	D $\rightarrow$ D $\uparrow$	ABC $\rightarrow$ D $\uparrow$	Spatial $\uparrow$	Object $\uparrow$	Goal $\uparrow$	Long $\uparrow$	Average $\uparrow$
MDT-V (FP16)	FP16	45.04 MB	4.52 $\pm$ 0.02	3.72 $\pm$ 0.06	77.9	87.5	75.4	65	76.5
w/ BatchQuant	FP8	22.52 MB	4.45 $\pm$ 0.10	3.65 $\pm$ 0.05	77.2	86.9	75.1	64.8	76.0
DC-QFA (Ours)			4.47 $\pm$ 0.07	3.67 $\pm$ 0.09	77.4	87.4	75.5	65.1	76.4
w/ SmoothQuant	W8A8	22.52 MB	4.23 $\pm$ 0.04	3.57 $\pm$ 0.09	72.8	81.9	71.5	62.1	72.0
w/ AWQ			4.21 $\pm$ 0.03	3.55 $\pm$ 0.06	72.8	82.1	72.2	62.9	72.5

Table 3: Results of MDT-V [ 25 ] on the CALVIN [ 21 ] and LIBERO [ 19 ] datasets, including performance metrics across various quantization methods and bit-widths. The table presents the model size, accuracy, and average performance (average length on CALVIN and success rate on LIBERO) for each method, highlighting the effectiveness of our DC-QFA when compared to other quantization methods.

Method	Bit-width	Model Size	CALVIN
Method	Bit-width	Model Size	D $\rightarrow$ D $\uparrow$	ABC $\rightarrow$ D $\uparrow$
MDT-V	FP16	45.04 MB	4.52 $\pm$ 0.02	3.72 $\pm$ 0.06
DC-QFA	FP8	22.52 MB	4.47 $\pm$ 0.07	3.67 $\pm$ 0.09
DC-QFA + OPD	FP8	22.52 MB	4.51 $\pm$ 0.04	3.72 $\pm$ 0.07
DC-QFA	W8A8	22.52 MB	4.48 $\pm$ 0.11	3.64 $\pm$ 0.07
DC-QFA + OPD	W8A8	22.52 MB	4.50 $\pm$ 0.07	3.69 $\pm$ 0.09

Table 4: Long-horizon evaluation on the CALVIN [ 21 ] benchmark using MDT-V [ 25 ] . We compare the quantized policies trained with DC-QFA and DC-QFA+OPD under FP8, W8A8, and W4A4 settings. Metrics report the average rollout length on the D

\rightarrow

D and ABC

\rightarrow

D splits (higher is better). While DC-QFA preserves most of the full-precision performance under quantization, incorporating OPD further improves rollout stability and consistently recovers performance closer to the FP16 baseline in long-horizon tasks.

Metric	OpenVLA-OFT (T1)	OpenVLA-OFT (T2)	DiffusionPolicy-T (T3)	DiffusionPolicy-T (T4)	MDT-V (T3)	MDT-V (T4)
Base Model (FP16)	18	14	11	13	13	14
AWQ (W4A4)	10	8	5	5	4	6
DC-QFA (W8A8)	18	14	10	13	11	12
DC-QFA (W4A4)	15	13	10	13	11	12

Table 5: Real-world manipulation results on the Inovo robotic platform under different quantization settings. We evaluate four tasks including Task 1: sweeping coffee beans into a dustpan , Task 2: picking up eggs and placing them into a box , Task 3: inserting a red pepper into a cup , and Task 4: picking up a carrot and placing it into a bowl . Each task is evaluated over 20 trials. We compare FP16 base model, a PTQ baseline using AWQ [ 18 ] with W4A4 quantization, W8A8 policies, and our DC-QFA W4A4 policies. Results show that DC-QFA significantly improves the robustness of low-bit policies in real-world manipulation tasks.

Figure 3: Pareto curves of performance-parameter trade-offs based on the MDT-V W8A8 setting.

Figure 4: For mixed-precision deployment, the bit-widths $b^{W}_{l}$ and $b^{A}_{l}$ for each layer are sampled from $\{\text{INT4},\text{INT8}\}$ . DC-QFA achieves superior performance compared to fixed-precision baselines.

Metric	FP16	W8A8	W4A4
Memory (GB)	15.2	7.9	4
Reduction	–	1.92 $\times$	3.80 $\times$
Latency (ms)	644.74	297.84	217.38
Speedup	–	2.16 $\times$	2.97 $\times$

Table 6: Performance, runtime memory, memory reduction, latency, and speedup under different quantization bit-widths on the Jetson Orin NX.

Hardware-Aware NAS

Hardware-aware neural architecture search selects and co-designs models for accuracy, latency, energy, memory, and target-device constraints.

See all 49 papers

ActQuant: Sub-4-bit Action-Guided Quantization for Vision-Language-Action Models
May 19, 2026Arash Akbari, Arman Akbari, Masih Eskandar +11Neural Network Quantization VLM Quantization
1Northeastern University · University of Georgia · 3Cisco Systems
RoboRouter: Training-Free Policy Routing for Robotic Manipulation
Mar 9, 2026Yiteng Chen, Zhe Cao, Hongjia Ren +9Robot Policy Learning Robotic Manipulation
South China University of Technology · Nanjing University · University of New South Wales +5
Refine Actions If Needed: Adaptive Action Chunk Refinement for Robot Manipulation
Oct 14, 2026Hanlin Liu, Zichen Zhou, Jiacheng Zhang +6Robot Policy Learning Robotic Manipulation
The Hong Kong University of Science and Technology (Guangzhou) · Beijing Institute of Technology · Shanghai Jiao Tong University +2

May 19, 2026J/K move · Enter open · S save

ActQuant: Sub-4-bit Action-Guided Quantization for Vision-Language-Action Models

Arash Akbari, Arman Akbari, Masih Eskandar, Qitao Tan+10

1Northeastern University · University of Georgia · 3Cisco Systems

Neural Network Quantization VLM Quantization Efficient VLA Model Inference Low-Bit Quantization Mixed-Precision Quantization Efficient VLA Models+4

Open paper AskPDF arXiv

Vision-Language-Action (VLA) models exhibit remarkable action generation for embodied intelligence, but their heavy compute make deployment on edge platforms impractical. Aggressive, sub-4-bit weight quantization is the natural solution, yet existing post-training quantization (PTQ) methods suffer severe performance degradation in this regime. To address this, we introduce ActQuant, an action-guided mixed-precision PTQ framework that operates in two stages: (1) an inter-tensor bit allocator that assigns each weight matrix a single bit-width based on how much it contributes to predicting the agent's actions; (2) an intra-tensor scale optimizer tunes per-block quantization scales using action-aware curvature, so that dynamic range is concentrated on the weights most influential for control. To deliver the on-device benefits of our aggressive quantization, we further introduce OmniModel.cpp, an agentic conversion pipeline that ports architectures into a native C/C++ runtime with efficient low-bit kernels. We evaluate ActQuant both in simulation and on a real-world 6-DoF UR3 arm, with all models deployed through OmniModel.cpp. On the LIBERO benchmark, ActQuant is the only method that operates at or below 3 bits-per-weight, retaining 95.0% on OpenVLA-OFT and 94.8% on

π_{0.5}

. Pushed further, ActQuant reaches 2.5 bpw at 90.1% on OpenVLA-OFT, compressing the backbone from 14.3 GB to 2.7 GB (5.3

\times

). On the physical UR3 arm,

π_{0.5}

quantized with ActQuant retains the baseline's success rate while reducing the memory footprint by 2.5

\times

Abstract

Hardware-Aware NAS

Explore similar work

ActQuant: Sub-4-bit Action-Guided Quantization for Vision-Language-Action Models

RoboRouter: Training-Free Policy Routing for Robotic Manipulation

Refine Actions If Needed: Adaptive Action Chunk Refinement for Robot Manipulation