cs.ROApr 14, 2026

E2E-Fly: An Integrated Training-to-Deployment System for End-to-End Quadrotor Autonomy

Authors: Fangyu Sun, Fanxing Li, Linzuo Zhang, Yu Hu, Renbiao Jin, Shuyu Wu, Wenxian Yu, Danping Zou

Organizations: School of Automation and Perception, Shanghai Jiao Tong University, China

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Training and transferring learning-based policies for quadrotors from simulation to reality remains challenging due to inefficient visual rendering, physical modeling inaccuracies, unmodeled sensor discrepancies, and the absence of a unified platform integrating differentiable physics learning into end-to-end training. While recent work has demonstrated various end-to-end quadrotor control tasks, few systems provide a systematic, zero-shot transfer pipeline, hindering reproducibility and real-world deployment. To bridge this gap, we introduce E2E-Fly, an integrated framework featuring an agile quadrotor platform coupled with a full-stack training, validation, and deployment workflow. The training framework incorporates a high-performance simulator with support for differentiable physics learning and reinforcement learning, alongside structured reward design tailored to common quadrotor tasks. We further introduce a two-stage validation strategy using sim-to-sim transfer and hardware-in-the-loop testing, and deploy policies onto two physical quadrotor platforms via a dedicated low-level control interface and a comprehensive sim-to-real alignment methodology, encompassing system identification, domain randomization, latency compensation, and noise modeling. To the best of our knowledge, this is the first work to systematically unify differentiable physical learning with training, validation, and real-world deployment for quadrotors. Finally, we demonstrate the effectiveness of our framework for training six end-to-end control tasks and deploy them in the real world.

Figures & tables

 TypeHardwareRenderSensorsAlgorithmValidationDeploymentNumber of Tasks
Flightmare [ 12 ]simulator×\timesUnityD,R,S,IRLHIL×\times×\times
RotorS [ 13 ]simulator×\timesOpenGLD,R,I×\times×\times×\times×\times
CrazyS [ 14 ]simulator×\timesOpenGLD,R,I×\times×\times×\times×\times
CrazySim [ 15 ]simulator×\timesOpenGLD,R,S,I×\times×\times×\times×\times
FlightGoggles [ 16 ]simulator×\timesUnityD,R,S,I×\timesHIL×\times×\times
FastSim [ 17 ]simulator×\timesUnityD,R,S,I,LRLHIL×\times×\times
TABLE I: Key characters comparison of state-of-the-art approaches in this area.
Fig. 1: The overview architecture of E2E-Fly. In the training phase, the state-based and vision-based inputs are acquired from VisFly. During this process, the reward function is designed according to the reward function manual, while the accurate dynamics model supports training both via RL and differentiable simulation. The trained policy can be directly transferred to AirSim via an internal interface for cross-platform sim-to-sim test. In the real-world deployment phase, we provide two quadrotor hardware platforms that support real-time hardware-in-the-loop simulation and onboard flight test, respectively. The transfer from policy outputs to low-level control commands is achieved through our open-source betaflight-ctrl package.
Alg.TaskReward
PPOHoveringrtprog+rtact+rtv+rtanglev+rtgoal+rtcrashr_{t}^{\text{prog}}+r_{t}^{\text{act}}+r_{t}^{\text{v}}+r_{t}^{\text{anglev}}+r_{t}^{\text{goal}}+r_{t}^{\text{crash}}
Landingrtprog+rtact+rtv+rtanglev+rtgoal+rtcrashr_{t}^{\text{prog}}+r_{t}^{\text{act}}+r_{t}^{\text{v}}+r_{t}^{\text{anglev}}+r_{t}^{\text{goal}}+r_{t}^{\text{crash}}
Trackingrtprog+rtact+rtori+rtv+rtanglev+rtgoal+rtcrashr_{t}^{\text{prog}}+r_{t}^{\text{act}}+r_{t}^{\text{ori}}+r_{t}^{\text{v}}+r_{t}^{\text{anglev}}+r_{t}^{\text{goal}}+r_{t}^{\text{crash}}
Racingrtprog+rtact+rtgoal+rtcrashr_{t}^{\text{prog}}+r_{t}^{\text{act}}+r_{t}^{\text{goal}}+r_{t}^{\text{crash}}
Obstacle avoidancertprog+rtori+rtv+rtanglev+rtavoid+rtgoal+rtcrashr_{t}^{\text{prog}}+r_{t}^{\text{ori}}+r_{t}^{\text{v}}+r_{t}^{\text{anglev}}+r_{t}^{\text{avoid}}+r_{t}^{\text{goal}}+r_{t}^{\text{crash}}
BPTTHoveringrtprog+rtact+rtori+rtv+rtanglevr_{t}^{\text{prog}}+r_{t}^{\text{act}}+r_{t}^{\text{ori}}+r_{t}^{\text{v}}+r_{t}^{\text{anglev}}
TABLE II: The reward design instruction for different types of tasks.
TaskCurriculum Design
High speed obstacle avoidanceC1: obstacle-free point navigation
C2: sparse obstacle navigation
C3: high-speed collision-free flight
Racing with random obstaclesC1: obstacle-free racing
C2: sparse obstacle racing
C3: dense obstacle racing
TABLE III: The examples of curriculum designs for complex tasks.
Fig. 2: The hardware-in-the-loop simulation in E2E-Fly. It consists of a real quadrotor flying in a motion capture system combined with a photorealistic simulation of complex 3D environments. Multiple sensors can be simulated with minimal delays while virtually flying in various simulated scenes. Such hardware-in-the-loop simulation offers a modular framework for prototyping robust vision-based algorithms safely, efficiently, and inexpensively.
Fig. 3: Proposed quadrotor platforms: VIS-R and VIS-H. VIS-R is used for onboard experiments with a Radax X4 and an Intel D435i RGB-D camera. VIS-H supports offboard and hardware-in-the-loop experiments via a wireless data transmitter and a wireless video transmitter.
Fig. 4: The digital prototype employed for moment of inertia test. The model is constructed by accurately assigning the material and mass properties for each component based on the physical quadrotor. Following this, the inertia tensor is computed within the software environment, enabling the extraction of the three principal moments of inertia directly from its diagonal.
Fig. 5: The LY-5KGF test stand for motor system identification. It‘s capable of test propellers of various sizes and featuring software support for remote monitoring and parameter tuning.
ParameterValue
mass mm [kg]0.470.47 for VIS-H, 0.750.75 for VIS-R
inertia J\bm{J} [kg ⋅\cdot m 2 ][ 1.251.25 , 1.281.28 , 2.032.03 ] ×10−3\times 10^{-3} for VIS-H, [ 1.411.41 , 1.531.53 , 2.052.05 ] ×10−3\times 10^{-3} for VIS-R
maximum thrust per motor fmax\bm{f}^{max} [N]5.125.12
arm length r\bm{r} [m]0.0750.075
motor time constant kmotork^{motor} [s]0.0350.035
thrust map [kf2,kf1,kf0][k_{f2},k_{f1},k_{f0}] [N ⋅\cdot s 2 /rad 2 , N ⋅\cdot s/rad, N][ 4.04×10−74.04\times 10^{-7} , 2.56×10−52.56\times 10^{-5} , −2.62×10−2-2.62\times 10^{-2} ]
TABLE IV: Accurate parameters of our hardware platforms
Task TypeTarget TaskRandomization Parameters
State-basedHoveringp,v\bm{p},\bm{v}
Landingp\bm{p}
Racingp\bm{p}
Trackingp,v\bm{p},\bm{v}
Vision-basedRacing with obstaclesp,pgate,qgate,pob\bm{p},\bm{p}_{\text{gate}},\bm{q}_{\text{gate}},\bm{p}_{\text{ob}}
Visual landingp,psize,pshape\bm{p},\bm{p}_{\text{size}},\bm{p}_{\text{shape}}
TABLE V: Domain randomization parameters for specific tasks.
Sensor TypeClasses 
VisionGaussian noise 
Salt & pepper noise 
Poisson noise 
Speckle noise 
Redwood depth noise 
ImuGaussian white noise 
TABLE VI: The type of noise provided in E2E-Fly [ 21 , 28 ] .
Fig. 6: Examples of benchmark training scenarios constructed in the E2E-Fly. We present the simulation performance across various tasks, demonstrating that all policies trained via differentiable simulation and RL can accomplish their objectives.
Fig. 7: Training rewards comparison between PPO and BPTT. It is evident from the figure that BPTT exhibits faster convergence, higher sample efficiency, and achieves a higher reward compared with PPO.
 FPS [its/s]time steps [its]Time [s]
 HoveringLandingTrackingRacingHoveringLandingTrackingRacingHoveringLandingTrackingRacing
PPO≈20000\approx 20000≈20000\approx 20000≈20000\approx 20000≈15000\approx 150005e71e85e71e8≈2500\approx 2500≈5000\approx 5000≈2500\approx 2500≈6600\approx 6600
BPTT≈13000\approx 13000≈13000\approx 13000≈24000\approx 24000≈20000\approx 200001e71e72e75e7≈750\approx 750≈750\approx 750≈800\approx 800≈2500\approx 2500
TABLE VII: We record the training FPS, the total time steps required to converge, and the overall training time for PPO and BPTT on the four baseline tasks. The FPS is measured with 100 parallel environments.
 ParameterValue
BPTToptimizerAdam
 learning rate1e-3 decay to 1e-5
 discount factor0.99
 horizon length96
 replay buffer size100000
 batch size25600
TABLE VIII: Parameters for policy training.
TaskObservationRewardExample Parameters
Hovering(pt,qt,vt,at)∈R13(\bm{p}_{t},\bm{q}_{t},\bm{v}_{t},\bm{a}_{t})\in\mathbb{R}^{13}rtprog2+rtori2+rtv1+rtanglev1r_{t}^{\text{prog2}}+r_{t}^{\text{ori2}}+r_{t}^{\text{v1}}+r_{t}^{\text{anglev1}} (D)λ2=0.01,λ10=0.0001,λ13=0.002,λ17=0.002\lambda_{2}=0.01,\lambda_{10}=0.0001,\lambda_{13}=0.002,\lambda_{17}=0.002
Landing(pt,qt,vt,at)∈R13(\bm{p}_{t},\bm{q}_{t},\bm{v}_{t},\bm{a}_{t})\in\mathbb{R}^{13}rtprogxy+rtvz+rtanglev1r_{t}^{\text{progxy}}+r_{t}^{\text{vz}}+r_{t}^{\text{anglev1}} (D)λ3=0.04,λ16=0.1,λ17=0.001\lambda_{3}=0.04,\lambda_{16}=0.1,\lambda_{17}=0.001
Tracking(pti=1−10,qt,vt,at)∈R40(\bm{p}_{t}^{i=1-10},\bm{q}_{t},\bm{v}_{t},\bm{a}_{t})\in\mathbb{R}^{40}rtprog2+rtori2+rtv1+rtanglev1r_{t}^{\text{prog2}}+r_{t}^{\text{ori2}}+r_{t}^{\text{v1}}+r_{t}^{\text{anglev1}} (D)λ2=0.02,λ10=0.001,λ13=0.002,λ17=0.002\lambda_{2}=0.02,\lambda_{10}=0.001,\lambda_{13}=0.002,\lambda_{17}=0.002
Racing(pt1,pt2,qt,vt,at)∈R16(\bm{p}_{t}^{1},\bm{p}_{t}^{2},\bm{q}_{t},\bm{v}_{t},\bm{a}_{t})\in\mathbb{R}^{16}rtprog1+rtori2+rtv1+rtanglev1r_{t}^{\text{prog1}}+r_{t}^{\text{ori2}}+r_{t}^{\text{v1}}+r_{t}^{\text{anglev1}} (D)λ1=0.9,λ10=0.001,λ13=0.002,λ17=0.002\lambda_{1}=0.9,\lambda_{10}=0.001,\lambda_{13}=0.002,\lambda_{17}=0.002
TABLE IX: The observation space and reward function of state-based tasks.
Fig. 8: Success rate and task error under different reward settings training via PPO. The figure presents the ablation study evaluating the effects of the individual reward components trained with PPO. The rfullr^{\text{full}} represents the configuration using all reward components listed in Table II , while w/o rvr^{\text{v}} ,w/o rprogr^{\text{prog}} ,w/o ractr^{\text{act}} ,w/o ranglevr^{\text{anglev}} ,w/o rsparser^{\text{sparse}} denote the cases without linear-velocity reward, progress reward, action-smoothness reward, angular-velocity reward, and sparse reward, respectively. For the racing task, only the rfullr^{\text{full}} , w/o rprogr^{\text{prog}} , w/o rsparser^{\text{sparse}} , w/o ranglevr^{\text{anglev}} reward curves are included.
Fig. 9: Success rate and task error under different reward settings training via BPTT. The figure presents the ablation study evaluating the effects of the individual reward components trained with BPTT. The rfullr^{\text{full}} represents the configuration using all reward components listed in Table II , while w/o rvr^{\text{v}} ,w/o rprogr^{\text{prog}} ,w/o ranglevr^{\text{anglev}} denote the cases without linear-velocity reward, progress reward, action-smoothness reward, angular-velocity reward, and sparse reward, respectively. For the racing task, only the rfullr^{\text{full}} and w/o rprogr^{\text{prog}} reward curves are included.
Fig. 10: Simulation result of vision-based landing. Subfigures (a), (b), and (c) depict the performance of the identical policy while landing on triangular, circular, and square landing pads, respectively. The corresponding segmentation maps from the downward-facing camera are displayed on the right side of each subfigure.
Fig. 11: Simulation result of racing with obstacles. Subfigures (a), (b), and (c) present simulation results obtained in S-shaped, 3D circle, and J-shaped race tracks, respectively, with maximum velocities achieving over 10 m/s.
 Visual landingRacing with obstacles
Observation(pt,qt,vt,at)∈R13(\bm{p}_{t},\bm{q}_{t},\bm{v}_{t},\bm{a}_{t})\in\mathbb{R}^{13} & segmentation∈R64×64{segmentation}\in\mathbb{R}^{64\times 64}(pt1,pt2,qt,vt,vd,at)∈R17(\bm{p}_{t}^{1},\bm{p}_{t}^{2},\bm{q}_{t},\bm{v}_{t},v_{d},\bm{a}_{t})\in\mathbb{R}^{17} & depth∈R64×64depth\in\mathbb{R}^{64\times 64}
Rewardrtprogxy+rtori2+rtvz+rtanglev1r_{t}^{\text{progxy}}+r_{t}^{\text{ori2}}+r_{t}^{\text{vz}}+r_{t}^{\text{anglev1}} (D)rtprog1+rtact3+rtori1+rtavoid1+rtgoal+rtcrashr_{t}^{\text{prog1}}+r_{t}^{\text{act3}}+r_{t}^{\text{ori1}}+r_{t}^{\text{avoid1}}+r_{t}^{\text{goal}}+r_{t}^{\text{crash}} (PD)
Parameterλ3=0.04,λ10=0.003,λ16=0.1,λ17=0.001\lambda_{3}=0.04,\lambda_{10}=0.003,\lambda_{16}=0.1,\lambda_{17}=0.001λ1=0.9,λ7=0.025,λ8=0.002,λ9=0.04,λ18=0.01,λ21=5,λ22=4\lambda_{1}=0.9,\lambda_{7}=0.025,\lambda_{8}=0.002,\lambda_{9}=0.04,\lambda_{18}=0.01,\lambda_{21}=5,\lambda_{22}=4
TABLE X: The observation space and reward function of vision-based tasks.
Fig. 12: Reward and success rate curves of visual tasks. From left to right, the figure depicts the reward and success rate curves for racing with obstacles and visual landing.
Fig. 13: The step-response alignment. We align the step-response of angular velocity and the mass-normalized thrust along the z-axis.
Fig. 14: Policy response after step-response alignment. The corresponding policy deployment without requiring additional alignment, yielding a response that closely matches the simulation.
Fig. 15: Zero-shot transfer from simulation to real world via our E2E-Fly system. Real-world performance demonstrates that the policy trained with our system can achieve zero-shot transfer from simulation to the real world.

Explore similar work

CardsList

  1. VisFly-Lab: Unified Differentiable Framework for First-Order Reinforcement Learning of Quadrotor Control

    Mar 22, 2026Fanxing Li, Fangyu Sun, Tianbao Zhang +5Reinforcement LearningQuadrotor Control

  2. Simple but Stable, Fast and Safe: Achieve End-to-end Control by High-Fidelity Differentiable Simulation

    Apr 12, 2026Fanxing Li, Shengyang Wang, Yuxiang Huang +5Quadrotor ControlVision-Based Robot Control

    Shanghai Jiao Tong University · Kunshan Duke University

  3. Efficient Knowledge Transfer for Jump-Starting Control Policy Learning of Multirotors through Physics-Aware Neural Architectures

    Feb 17, 2026Welf Rehberg, Mihir Kulkarni, Philipp Weiss +1Robotic RLCross-Embodiment Robot Learning

    Department of Engineering Cybernetics at the Norwegian University of Science and Technology, O.S. Bragstads Plass 2D, 7034, Trondheim, Norway