MobiExo: GPS-SLAM Fusion for Seamless Indoor-Outdoor Mobile Manipulation with Hand-Foot Coordination

Jianpeng Wang, Zhen Tian, Wenlong Chen, Dian Yuan, Zhou Zhou, Cen Ming, Xia Hua, Fei YuIROS 2025 · OpenReview

TL;DR

MobiExo is an innovative teleoperation system that enables seamless indoor-outdoor mobile robot manipulation by combining a wearable exoskeleton with an advanced sensor fusion framework. By unifying GPS and SLAM through a self-adaptive federated filter and introducing an intuitive hand-foot coordination mapping, the system eliminates localization transition errors and allows operators to control complex robot movements with millimeter-level precision.

What Problem Does This Paper Address?

While mobile robots equipped with robotic arms hold immense promise for logistics, maintenance, and field robotics, teleoperating them across diverse environments remains a massive headache. Current systems suffer from two core limitations:

  1. The Localization Chasm: Indoor tracking typically relies on Visual-Inertial SLAM, while outdoor tracking relies on GPS. When a robot crosses a threshold (like a doorway), switching abruptly between these modalities causes massive "position jumps" (often exceeding 80 cm), which can cause the robot to crash or drop objects.
  2. Decoupled Control Interfaces: Most mobile manipulators force the operator to split their attention—using a joystick to drive the base and an exoskeleton or digital twin interface to move the arms. This high cognitive load makes fluid, simultaneous moving-and-manipulating nearly impossible.

MobiExo tackles these challenges head-on to unlock continuous, whole-body mobile teleoperation.

Methods

The MobiExo architecture bridges the gap between human intent and robot execution through a tightly integrated hardware and software stack.

Fig. 2: Detailed system architecture of MobiExo showing both operator and robot components. The operator wears a backpack with battery, GPS/IMU modules, and computing unit connected to the ACE exoskeleton. The mobile robot incorporates processing units, sensors, and dual manipulators, communicating wirelessly with the operator system.

1. Self-Adaptive Federated Filter

Instead of hard-switching between SLAM and GPS, MobiExo uses a master-local filter architecture. It continuously calculates environment-specific "reliability scores" for each sensor based on metrics like Signal-to-Noise Ratio (SNR) and satellite count for GPS, or tracked features and reprojection errors for SLAM.

Fig. 3: MobiExo hybrid localization system with self- adaptive federated filter architecture. The system processes measurements from multiple sources through local filters, adapting information distribution based on sensor reliability.

When transitioning between zones, the system utilizes a sigmoid-based soft-switching function to smoothly shift the weights of the covariance matrices, preventing sudden localization spikes.

2. Hand-Foot Coordination Framework

To make control intuitive, the system maps the operator's upper-body movements directly to dual 6-DOF manipulators via a 7-DOF wearable exoskeleton. Simultaneously, lower-limb inertial sensors capture the operator's natural gait primitives (walking forward, stepping laterally, rotating) at 200Hz to drive the robot's omnidirectional mobile base.

Crucially, an Adaptive Gain Control algorithm dynamically scales down manipulation sensitivity during rapid walking to maintain stability, and scales it back up for high-precision tasks when the operator stands still.

Datasets

The paper does not train on a standard public machine learning dataset; instead, it introduces a robust multimodal data collection pipeline built into the ROS2 Humble architecture. While the operator teleoperates the robot, the system automatically records time-synchronized data for imitation learning:

  • Kinematic Data (100Hz): Joint positions, velocities, and torques.
  • Localization Data: GPS coordinates and SLAM estimates.
  • Visual Observations: RGB and depth images (7–18 GB/hour).

The system was evaluated using extensive field trials across a custom mixed environment featuring office corridors, distinct transition entryways (sliding doors, loading docks), and varied outdoor terrains (gravel, grass, slopes).

Key Insights and Ideas

  • Information-Level Fusion is Key to Smoothness: Treating localization transitions as a continuous spectrum rather than a binary switch reduces boundary errors by roughly 74%.
  • Gait-Aware Manipulation: Humans naturally stabilize their hands when walking. By mimicking this via the Adaptive Gain Control metric (derived from the operator's center-of-mass velocity), the robot successfully dampens the physical vibrations caused by its mobile base during locomotion.
  • Lowering the Barrier to Entry: Traditional joystick-plus-exoskeleton setups require significant training (12–15 trials). By mapping locomotion to natural walking patterns, operators achieved proficiency in just 3–4 trials, proving that full-body coordination drastically reduces cognitive load.

Key Results

MobiExo significantly outperformed baseline teleoperation methods across multiple metrics:

  • Localization Accuracy: Achieved continuous positioning errors of just 4.5±0.8 cm indoors (SLAM) and 6.8±1.2 cm outdoors (RTK-GPS).
  • Transition Performance: During tricky indoor-outdoor crossings, MobiExo kept localization errors under 15 cm, avoiding the >80 cm jumps typical of standard switching methods.
Fig. 4: Indoor-outdoor transition sequence showing the federated filter operation across three phases. The system maintains continuous position tracking with minimal error spikes (green) compared to traditional switching methods (red dashed line).
  • Task Success & Efficiency: In multi-location assembly tasks, MobiExo demonstrated a 5.9× efficiency improvement over stationary setups (which require constant repositioning/recalibration) and a 2.5× improvement over joystick-controlled mobile bases. It achieved an 87.5% success rate during transitions, compared to just 41.3% for the joystick baseline.
  • Precision: Maintained an end-effector precision of 5.8±0.7 mm even while the robot was actively moving.

Limitations and Future Work

Despite its impressive performance, the authors note a few key constraints:

  1. Feature-Poor, GPS-Denied Spaces: In long, featureless concrete or metal corridors where GPS signals are blocked and SLAM has no visual anchors to track, the system reverts to IMU dead reckoning, resulting in a position drift of ~2.3 cm/s.
  2. Computational Bottlenecks: Simultaneous tracking, obstacle avoidance, and manipulation planning caused processing delays (50–100ms) during roughly 5.2% of operational testing, primarily driven by heavy SLAM processing during rapid movements.
  3. Communication Vulnerabilities: Heavy radio-frequency (RF) interference in industrial settings occasionally caused data dropouts.

Future iterations aim to integrate semi-autonomous behaviors (like automated obstacle avoidance) to further alleviate operator workload and mitigate compute bottlenecks.

Summary updated

Abstract

Teleoperation systems for mobile robots face significant challenges in achieving seamless coordination across dynamic environments. We present MobiExo, a teleoperation system that unlocks seamless indoor-outdoor mobile manipulation. Our approach tackles two fundamental challenges: robust cross-environment localization and intuitive full-body control. A novel self-adaptive federated filter unifies GPS and SLAM, delivering continuous centimeter-level positioning (4.5±0.8 cm indoor, 6.8±1.2 cm outdoor) and eliminating transition errors. Simultaneously, an integrated hand-foot coordination framework translates the operator’s natural gait and gestures into fluid robot actions, maintaining remarkable millimeter-level end-effector precision (3.5±0.4 mm) during navigation. Extensive field trials validate our design, demonstrating high task success (96.7% indoor, 94.3% outdoor) and a 5.9× efficiency improvement in multi-location tasks over stationary setups. Code is available at: https://github.com/wangjianpeng200/MobiExo.git
BibTeX
@misc{wang2025mobiexo,
  author = {Jianpeng Wang and Zhen Tian and Wenlong Chen and Dian Yuan and Zhou Zhou and Cen Ming and Xia Hua and Fei Yu},
  title = {{MobiExo: GPS-SLAM Fusion for Seamless Indoor-Outdoor Mobile Manipulation with Hand-Foot Coordination}},
  year = {2025},
  url = {https://openreview.net/forum?id=acTlh7SJRV},
  howpublished = {IROS},
}
Figure