C2Flight: A Closed-Loop Co-Simulation Framework for Early-Stage Evaluation of Heterogeneous UAV Computing Systems

Xizhu Chen; Shuyuan Zhang; Yixuan Li; Huazhong Yang; Yu Wang; Jincheng Yu; Hongyang Jia

cs.ROOct 10, 2026

C2Flight: A Closed-Loop Co-Simulation Framework for Early-Stage Evaluation of Heterogeneous UAV Computing Systems

Authors: Xizhu Chen, Shuyuan Zhang, Yixuan Li, Huazhong Yang, Yu Wang, Jincheng Yu, Hongyang Jia

Abstract

Early-stage design of heterogeneous UAV computing systems requires evaluating how algorithm, hardware, communication, and system topology affect closed-loop flight before the complete system is available. However, design choices interact through the robotic feedback loop, making task performance difficult to predict from isolated component metrics, while existing execution-driven approaches postpone system-level design feedback until the computing system is integrated. This work presents C2Flight, a closed-loop co-simulation framework that represents target systems as parameterized dataflow graphs and integrates models of hardware under design with independently accessible physical hardware through a unified node interface. Its graph runtime captures the combined effects of buffering, activation, resource sharing, communication, and local timing on closed-loop flight behavior. Validation against onboard and wireless offboard physical computing pipelines confirms its timing fidelity and closed-loop prediction accuracy. Two case studies demonstrate early-stage exploration of deployment and accelerator bit-width choices. These results show that C2Flight exposes system-level design trade-offs before full-system integration, enabling rapid early-stage design refinement.

Figures & tables

Fig. 1: Early-Stage Design of UAV Computing Systems with C2Flight. Candidate algorithms, compute placement, communication links, and hardware configurations are represented by a parameterized graph, while a unified interface combines component models with available hardware for closed-loop evaluation. The resulting system-level feedback supports candidate comparison, bottleneck identification, and optimization prioritization, closing the design feedback loop before full-system integration.

Fig. 2: Overview of C2Flight. The graph exposes computation and interaction boundaries. GraphConfig defines inter-node policies, while NodeConfig selects implementations and local timing. The unified interface connects these descriptions to GraphRuntime and the environment adapter. Experimental access overhead is recorded separately and excluded from target latency where the execution contract permits this separation. Compatible replacements preserve declared port semantics and graph policies for system-level comparison. Solid blue arrows indicate data exchange; dashed blue arrows indicate runtime coordination and backend binding.

Group	Parameter	Representative choices
Data exchange $(\Theta_{d})$	Port storage mode	Queued storage, latest-value storage, direct handoff
	Full-buffer policy	Blocking, dropping new inputs, dropping oldest inputs
	Read policy	Data consumption, data retention
	Packet selection policy	FIFO order, latest arrival, latest source timestamp
	Connection mode	Direct transfer, broadcast, tagged routing
Node activation $(\Theta_{a})$	Trigger type	Data-driven, periodic, completion-triggered, external

TABLE I: Representative choices for the three graph interaction groups and their shared time convention. Node implementations, local timing sources, and access settings belong to node configurations.

Fig. 3: Three backend classes under the unified node execution interface. (a) call returns locally computed outputs. (b) request obtains device outputs and optional measurements. For both classes, the runtime combines local timing with graph policies to determine target events. (c) timed reports controlled boundary events, receives task-level transfer authorizations, and confirms common-boundary progress. The classes share port-event meanings but differ in timing and state-control responsibilities.

Onboard					Offboard
	Mean	Std.	$W_{1}\downarrow$	Norm. $W_{1}$ (%)		Mean	Std.	$W_{1}\downarrow$	Norm. $W_{1}$ (%)
CH	127.76	11.51	3.02	2.42	CH	415.50	34.33	7.82	1.88
CM	127.72	11.51	2.98	2.39	CM	415.87	34.28	7.61	1.83
SM	121.37	0.01	10.28	8.24	SM	226.05	18.99	165.21	42.22
GT	124.74	11.54	–	–	GT	416.86	35.89	–	–

TABLE II: Command-age statistics and prediction errors

Onboard
Speed (m/s)	GT	CH	CM	SC	SW
0.80	7/7	7/7	7/7	7/7	7/7
0.90	7/7	7/7	6/7	7/7	7/7
1.00	2/7	1/7	0/7	7/7	7/7
1.10	1/7	0/7	0/7	5/7	7/7
1.20	0/7	0/7	0/7	0/7	0/7

TABLE III: Task-success rates across the speed sweep

Config.	Position MAE (cm)	Integrated error (m $\cdot$ s)
CH	3.040	0.326
CM	3.782	0.406
SC	12.456	1.335
SW	14.661	1.570

TABLE IV: Onboard trajectory errors at 1.0 m/s, averaged over seven paired runs relative to GT.

Fig. 4: Illustrative mean trajectories for the onboard system at 1.0 m/s. Errors in Table IV are computed from paired runs, not between mean curves.

Fig. 5: Onboard–offboard deployment exploration. Left: effective new-result rate and inverse command-ready latency, corresponding to the reciprocals of the control update period and end-to-end latency. Right: highest tested speed cap with at least one completed trial.

Speed cap	INT4	INT8	FP16	FP32
$0.50$ m/s	5/5	5/5	5/5	5/5
$0.75$ m/s	5/5	5/5	4/5	2/5
$1.00$ m/s	0/5	5/5	0/5	0/5
$1.25$ m/s	0/5	0/5	0/5	0/5

TABLE V: Accelerator precision selection using CM

Heterogeneous Computing

Heterogeneous computing studies how to coordinate different processors, accelerators, and memory systems to improve performance, efficiency, and cost.

See all 80 papers

C $^2$ -Explorer: Contiguity-Driven Task Allocation with Connectivity-Aware Task Representation for Decentralized Multi-UAV Exploration
Mar 8, 2026Xinlu Yan, Mingjie Zhang, Yuhao Fang +5Multi-Robot Exploration Multi-Robot Task Allocation
School of Intelligence Science and Engineering, Harbin Institute of Technology, Shenzhen, Guangdong, China · Robotics and Autonomous Systems Thrust, The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, Guangdong, China · Department of Mechanical and Energy Engineering, Southern University of Science and Technology, Shenzhen, Guangdong, China
SwarmFly: A simulation platform for UAV swarm experiment design and validation
Jun 23, 2026Abhishek Phadke, Karthik Kumar Vasudeva, Abhishek JoshiAerial Robotics Swarm Robotics
School of Engineering and Computing, Christopher Newport University · Department of Mathematics and Statistics, Texas Tech University · Department of Computer Science, Texas A&M University-Corpus Christi
Robust Co-design Optimisation for Agile Fixed-Wing UAVs
Mar 11, 2026Adrian Andrei Buda, Xavier Chen, Nicolò Botteghi +1Robot Morphology-Control Co-Design Aerial Robotics
Department of Aeronautics, Imperial College London, United Kingdom · Department of Mathematics, Politecnico di Milano, Italy

Mar 8, 2026J/K move · Enter open · S save

C $^2$ -Explorer: Contiguity-Driven Task Allocation with Connectivity-Aware Task Representation for Decentralized Multi-UAV Exploration

Xinlu Yan, Mingjie Zhang, Yuhao Fang, Yanke Sun+4

School of Intelligence Science and Engineering, Harbin Institute of Technology, Shenzhen, Guangdong, China · Robotics and Autonomous Systems Thrust, The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, Guangdong, China · Department of Mechanical and Energy Engineering, Southern University of Science and Technology, Shenzhen, Guangdong, China

Multi-Robot Exploration Multi-Robot Task Allocation Multi-Robot Coordination

Open paper AskPDF arXiv

Efficient multi-UAV exploration under limited communication is severely bottlenecked by inadequate task representation and allocation. Previous task representations either impose heavy communication requirements for coordination or lack the flexibility to handle complex environments, often leading to inefficient traversal. Furthermore, short-horizon allocation strategies neglect spatiotemporal contiguity, causing non-contiguous assignments and frequent cross-region detours. To address this, we propose C

^2

-Explorer, a decentralized framework that constructs a connectivity graph to decompose disconnected unknown components into independent task units. We then introduce a contiguity-driven allocation formulation with a graph-based neighborhood penalty to discourage non-adjacent assignments, promoting more contiguous task sequences over time. Extensive simulation experiments show that C

^2

-Explorer consistently outperforms state-of-the-art (SOTA) baselines, reducing average exploration time by 43.1% and path length by 33.3%. Real-world flights further demonstrate the system's feasibility. The code will be released at https://github.com/Robotics-STAR-Lab/C2-Explorer

Abstract

Heterogeneous Computing

Explore similar work

C2^22-Explorer: Contiguity-Driven Task Allocation with Connectivity-Aware Task Representation for Decentralized Multi-UAV Exploration

SwarmFly: A simulation platform for UAV swarm experiment design and validation

Robust Co-design Optimisation for Agile Fixed-Wing UAVs

C $^2$ -Explorer: Contiguity-Driven Task Allocation with Connectivity-Aware Task Representation for Decentralized Multi-UAV Exploration