Ground-in-the-Loop (GITL) in Space Robotics

• GITL is an operational paradigm that integrates ground-based human control and hardware to correct localization and support autonomous space systems.
• GITL simulation facilities employ hardware-in-the-loop testbeds, combining sensors like LIDAR and stereo cameras with nested feedback loops for realistic mission rehearsals.
• Advanced control methods such as UKF, impedance/admittance control, and ICP-based alignment in GITL significantly improve maneuver precision and mitigate operational risks.

Ground-in-the-Loop (GITL) denotes operational paradigms and simulation facilities in which ground-based human operators or hardware play a decisive role within the control or corrective feedback loops of robotic or autonomous space systems. GITL architectures are critical for bridging autonomy gaps during planetary exploration, on-orbit servicing (OOS), and active debris removal (ADR), especially where real-time decision-making and high-confidence state estimation are required but onboard sensing or processing are insufficient. GITL is distinguished from pure Earth-in-the-Loop (where human commands drive high-level actions) and full autonomy (where all control and state estimation are onboard).

1. GITL in Absolute Localization for Planetary Rovers

GITL localization has long represented the default approach for recovering the absolute position of near-Earth planetary rovers relative to orbital maps. In this paradigm, rovers estimate their state via chained wheel odometry, visual odometry, and inertial measurements, accumulating typical drift on the order of 2%. Once relative errors threaten to exceed mission-specific thresholds (e.g., 10 m absolute error), a GITL cycle is triggered. This engenders the following loop: (1) recent rover maps or images are downlinked to Earth, (2) ground operators visually register these against high-resolution orbital imagery or DEMs, and (3) the corrected absolute pose is uplinked to the rover (Cauligi et al., 2023). The human-in-the-loop process is both a throughput limiter and a navigation risk, causing daily traverse distances to be capped around O(10^2 m) and increasing mission dependence on Earth-based operations.

2. GITL Simulation Architectures in Space Robotics

On-ground GITL simulation facilities are essential for integrated development, validation, and demonstration of OOS, ADR, and docking maneuvers. One notable hardware-in-the-loop (HiL) testbed comprises two 6-DOF robotic arms (UR5e and UR10), stereo and depth cameras (Intel RealSense D415, ZED), LIDAR, and force/torque sensors. The facility is structured around tightly coupled nested feedback loops operating at 100 Hz: an outer software-based orbit and dynamics simulation loop (SGP4 orbit propagator, PyBullet for 6-DOF dynamics), and an inner hardware loop interfacing ROS controllers, robotic arms, and end-effectors. This configuration enables closed-loop, physics-informed emulation of LEO rendezvous, tumbling motion, and cooperative/captive docking (Sah et al., 2023).

Architectural Components of a Typical GITL Robotic Facility

Component	Model/Type	Function
Chaser Arm	UR5e + 7-axis rail	Manipulator, docking probe
Target Arm	UR10 (floor-fixed)	Target, simulates satellite
Main Sensors	RealSense D415, LIDAR, ZED	Pose/velocity estimation
End-effectors	RG2/RG2-v2 grippers	Capture, apply forces
Dynamics Simulation	SGP4 + PyBullet	Orbit propagation, contact

3. Signal Processing and Control Methodologies in GITL

State estimation and admittance/impedance control are central to both hardware and localization GITL implementations. For OOS/ADR simulation, sensor data streams from stereo cameras and LIDAR are fused via point cloud processing (PCL), outlier rejection (Statistical-RANSAC), and ICP-based template alignment to derive 6-DOF target poses. Tumbling motion is characterized via ZED stereo+IMU and smoothed by an Unscented Kalman Filter (UKF). Grasping routines and reactive compliance are realized using impedance or admittance control laws:

$$M_d (\ddot{\mathbf e}) + B_d (\dot{\mathbf e}) + K_d (\mathbf e) = \mathbf F_{\rm cmd} - \mathbf F_{\rm meas}$$

or equivalently in admittance form:

$$\dot{\mathbf x}_{\rm des} = M_a^{-1}\big(\mathbf F_{\rm meas} - B_a \dot{\mathbf x}_{\rm meas}\big)$$

where $\mathbf{e}$ is the error between desired and measured pose. The orbit and dynamics loop emulates 6-DOF rigid body motion under manipulator contact:

$$M_t \ddot{\mathbf x} + C_t(\dot{\mathbf x},\mathbf x)\,\dot{\mathbf x} + K_t\,\mathbf x = \mathbf F_c$$

4. Limitations and Bottlenecks of GITL

GITL-based absolute localization for planetary rovers imposes stringent limitations. Each GITL cycle is constrained by communication bandwidth, latency, and availability of ground operators, bottlenecking daily traverse to a few hundred meters (Cauligi et al., 2023). Missed communication windows or extended delays allow odometry drift to accumulate, potentially breaching mission requirements for absolute error (e.g., when $\|\mathrm{error}\| > 10$ m). In on-orbit scenarios, the fidelity of the simulation and alignment between real motion and commanded trajectories (RMS error $<$ 8 mm, force-tracking error $\sim$ 0.6 N) are empirically validated in hardware-in-the-loop GITL facilities (Sah et al., 2023). The continued reliance on GITL further hinders continuous autonomous operation, introduces operational risk, and increases total mission cost.

5. Automation and Alternatives: Towards GITL-Free Operations

ShadowNav exemplifies developments seeking to obviate GITL cycles through real-time, autonomous, crater-based localization. This system replaces human-in-the-loop absolute pose corrections with a particle filter that performs crater edge detection (using stereo discontinuities, Canny-shadow hybrids, HED networks) and matches detected rim points against orbital crater maps. The Q-score quantifies the match:

$$Q_{\mathrm{inc}} = \epsilon + \sum_{j=1}^m \|z_{j,world} – c_{nearest}\|_2$$

$$Q(b_i^t) = \min \left(1, (Q_{\mathrm{inc}}/m)^{-1} \right)$$

Monte Carlo simulations on synthetic lunar traverses demonstrate absolute localization errors $<$ 2 m, well within standard mission-level requirements, and eliminate the need for periodic GITL resets (Cauligi et al., 2023). A plausible implication is that robust, perception-driven onboard localization exclusively using high-frequency stereo and active illumination is feasible for multi-kilometer traverses in darkness, permanently shadowed regions, or light-constrained environments.

6. Practical Impact and Validation Metrics

GITL facilities and methodologies allow gradual transition toward higher autonomy by providing ground-truth validation environments and augmenting the fidelity of guidance, navigation, and control (GNC) pipelines for spacecraft and robots. Empirical results from on-ground HiLS for OOS/ADR demonstrate:

• Positioning accuracy (chaser vs. commanded): RMS $<$ 8 mm; peak $<$ 15 mm
• Force-tracking error: $\sim$ 0.6 N (mean absolute), stdev $<$ 1 N
• Pose estimation (ICP+LIDAR): 6-DOF pose error $<$ 3 mm/$<$ 0.5°
• Grasp success rate: 93% over 30 trials
• End-to-end control loop latency: $\sim$ 25 ms (Sah et al., 2023)

Similarly, crater-based autonomous localization successfully meets mission error bounds ($\|\mathrm{error}\|\!\!<\!10$ m with typical performance $\approx\!2$ m), while obviating operational bottlenecks and costs associated with GITL (Cauligi et al., 2023).

7. Future Directions and Operational Considerations

Operational constraints and recent advances suggest a migration from human-in-the-loop operational paradigms to fully autonomous, perception-informed navigation and servicing. The continued refinement of hardware-in-the-loop simulation testbeds, extensive use of GITL for mission rehearsal, and the empirical maturation of deep learning-based detection and filtering methods are converging to reduce dependence on ground operators. However, in scenarios where uncertainties are high or failures intolerable, GITL retains a critical safety and validation role, providing a final corrective capability when onboard systems cannot guarantee stringent mission-level performance.

---

Key References:

• "ShadowNav: Crater-Based Localization for Nighttime and Permanently Shadowed Region Lunar Navigation" (Cauligi et al., 2023)
• "Development of On-Ground Hardware In Loop Simulation Facility for Space Robotics" (Sah et al., 2023)