Stinger Robot: Autonomous Underground Drilling

• Stinger Robot is a compact, self-bracing platform with a mechanically self-locking tri-leg system designed for high-force drilling in confined, irregular hard-rock mines.
• It employs a force-aware, closed-loop control strategy using a four-phase finite-state machine implemented in ROS 2 to enable adaptive load sharing and precise positioning.
• The design overcomes conventional drilling limitations by safely transferring loads and maintaining stable anchorage in treacherous, infrastructure-less underground environments.

Stinger Robot is a compact, self-bracing robotic platform designed for autonomous, high-force drilling in confined, irregular, infrastructure-less underground mines, especially abandoned hard-rock mines that are too narrow or unstable for conventional drill rigs. Its defining integration is a mechanically self-locking tri-leg bracing mechanism with force-aware, closed-loop control and a front-mounted drilling unit, implemented as a four-phase finite-state machine in ROS 2. The platform is positioned as the first validated robotic architecture to combine distributed force bracing and autonomous drilling specifically for hard-rock underground environments with no supporting infrastructure (Liu et al., 31 Jul 2025).

1. Operational setting and system role

The Stinger Robot addresses a specific gap in underground mining robotics: abandoned, deep underground mines often have narrow, irregular cross-sections, unstable terrain, poor ground support, and no installed infrastructure. These conditions are incompatible with conventional drilling machinery, which is typically large and monolithic, depends on flat, stable ground and large open drifts, and achieves drilling stability through heavy hydraulic stabilizers anchored into the floor or walls. In collapsed or constricted tunnels, such machines cannot be operated safely or at all (Liu et al., 31 Jul 2025).

Within this context, the Stinger is defined not as a general-purpose mining vehicle but as a compact drilling anchor platform. Its three primary functions are robust self-anchoring to tunnel walls using a tri-leg bracing mechanism, safe load transfer of drilling forces and vibrations to ground without overloading joints, gearboxes, or support structures, and accurate positioning of a drilling unit mounted on its body. This division of roles is central to the larger modular mining concept described for the system: exploration, deployment, drilling, and support can be distributed across multiple robotic agents rather than concentrated in a single machine.

A common misconception is to treat the platform as a scaled-down substitute for a conventional drill rig. The design intent is narrower and more specialized. It is optimized for autonomous high-force drilling in confined spaces where external frames, flat platforms, rails, or foundations are unavailable, and where anchoring must occur directly against irregular tunnel surfaces.

2. Mechanical architecture and self-locking bracing

The robot consists of a compact central body and three deployable legs, termed “stingers.” The body houses electronics, including a companion computer, microcontroller, and power module, provides mechanical interfaces for a front-mounted drill unit, and serves as the reference frame for the legs. Of the three legs, the central leg has fixed orientation with linear motion only, while each side leg combines a revolute joint for angular deployment and a prismatic actuator for axial extension. The total number of actuated degrees of freedom in the bracing system is five: three prismatic and two revolute (Liu et al., 31 Jul 2025).

The drill is modeled as a front-facing unit attached to the body in simulation and envisioned in hardware. It includes a revolute joint for alignment and a prismatic joint for extending the bit into the rock. This architecture places the central leg approximately on the drilling axis, so that the dominant axial reaction is transferred through the leg best aligned to sustain it, while the side legs provide lateral constraints and reject reaction torques and off-axis bending moments.

The defining mechanical concept is “mechanically self-locking” tri-leg bracing. Once deployed and loaded, the worm-gear-based rotary joints of the side legs cannot be backdriven by reaction loads, and the linear actuators are also non-backdrivable within their nominal load range. The result is a braced configuration that maintains contact forces and geometry even if power is cut and that can absorb large drilling forces without continuous active torque. Self-locking is therefore structural rather than merely control-dependent.

The actuator and drive selections reflect this emphasis on preload and retention. Each leg uses a VEVOR linear actuator with total length 625 mm, stroke 500 mm, and maximum pushing force 1000 N, with extension constrained by

$$l_i \in [l_{\min}, l_{\max}], \quad l_{\min} = 625 \text{ mm},\; l_{\max} = 1125 \text{ mm}.$$

Each side rotary joint uses a two-stage reduction, comprising a 1:2 spur gear stage and a 1:40 worm gear stage, for an overall ratio of $R = 80{:}1$. The drive motor is a DFROBOT 12 V metal DC motor with stall torque $T_{in} \approx 1.765\,\text{Nm}$. With estimated overall efficiency $\eta \in [0.38, 0.882]$, the paper reports a conservative output torque estimate of $T_{out} \approx 53.656\,\text{Nm}$, sufficient to rotate the side legs into position under load while remaining non-backdrivable.

The electronics stack is correspondingly compact. High-level computation runs on a LattePanda 3 Delta under ROS 2. Low-level control runs on an Arduino Mega 2560 Pro communicating via USB serial, driving actuators through Pololu G2 18v17 motor drivers, reading encoder signals on rotary motors, and monitoring rotational limit switches as safety stops. Power is supplied by a DFR0205 power module providing a 12 V regulated supply.

3. Kinematics, workspace, and contact mechanics

The kinematic analysis is formulated in a global coordinate frame located at the central leg’s anchor point on the wall. The side-leg rotations satisfy

$$\theta_i \in [\theta_{\min}, \theta_{\max}], \quad \theta_{\min} = -90^\circ,\; \theta_{\max} = 90^\circ.$$

In the paper’s simplified two-dimensional analysis, the central leg lies on the body axis and the side legs sweep arcs as $\theta_i$ changes, with radial reach determined by $l_i$. This immediately yields a workspace dependence not only on actuator limits but also on tunnel geometry:

$$\text{Workspace} = f(l_{\min}, l_{\max}, \theta_{\min}, \theta_{\max}, \text{tunnel geometry}).$$

The analysis shows a characteristic regime structure. At minimum or maximum extension for all legs, the body has essentially a single reachable pose for a given tunnel cross-section. At intermediate extensions, the circular reachable regions of the legs overlap, creating a finite workspace region in which the body can be repositioned. If the tunnel is too large, the legs must fully extend to reach the walls and the workspace shrinks to a point. If the tunnel is too narrow, body-wall interference limits or eliminates repositioning. This suggests that Stinger is best understood as a geometry-conditioned anchoring platform rather than a universally deployable tunnel robot.

The tri-leg contact structure forms a closed kinematic chain between body and tunnel surface once all three legs are in contact. Three anchor points are treated as the minimum needed to determine a stable plane and define a rigid connection to the environment. The central leg carries the primary axial load because it is aligned with the intended drilling axis, while the side legs provide lateral stabilization and moment resistance. The configuration is therefore simultaneously a geometric support structure and a load-path design.

Although the paper does not explicitly derive full frictional contact models, it states the classical no-slip condition for each leg contact:

$F_{t,i} \le \mu_i F_{n,i}.$

Equivalently, the net contact force must remain within the friction cone. The hard bracing phase increases the normal force $R = 80{:}1$0 so that expected tangential loads from drilling, gravity, or other disturbances remain admissible. The mechanical interpretation is straightforward: anchoring quality depends not only on leg force capacity but also on rock-contact friction, which can vary significantly in real mines.

Simulation results show a consistent force-sharing pattern across mission phases. After initial bracing, the side legs each carry about 10 N and the central leg about 20 N, corresponding to an approximately 50/25/25 load split. After hard bracing, the side legs rise to about 140 N each and the central leg to about 280 N, preserving the same approximate distribution. During drilling, the central leg rises to about 1000 N, the side legs to about 273 N and 261 N, and the drill bit contact force to about 239 N. This confirms the intended axial dominance of the central leg and the stabilizing role of the side pair.

4. Force-aware control and ROS 2 implementation

The control strategy is explicitly force-aware and closed-loop. In simulation, and as planned for hardware, force sensors are placed at each leg tip and at the drill bit, with rotary-joint encoders and actuator state information supplied through ROS 2 Control and the Arduino interface. Force readings $R = 80{:}1$1 are monitored against thresholds to detect initial contact, regulate preload during bracing, and detect excessive loads during drilling (Liu et al., 31 Jul 2025).

Control is organized as a four-phase finite-state machine. In the opening phase, the side legs rotate into an approximate deployment configuration under position control, typically forming a Y shape. In the initial bracing phase, all three prismatic actuators extend outward under velocity control until each leg’s force sensor reaches the contact threshold, at which point that leg stops. In the hard bracing phase, the legs continue extending at reduced speed until the hard-bracing threshold is reached and maintained for a dwell time, after which each leg stops further extension. In the drilling phase, the drill revolute joint aligns the drilling axis, the drill prismatic joint advances until drill-bit contact is detected, drill rotation begins, and structural loads are continuously monitored; if any leg reaches the maximum safe force threshold or other safety limits are exceeded, drilling extension and/or rotation is stopped.

A central property of this controller is that extension terminates independently for each leg once the relevant threshold is met. The final bracing geometry is therefore environment-adaptive: irregular tunnel surfaces cause some legs to contact earlier and carry slightly different loads, while the controller enforces force-based rather than purely position-based deployment. The paper identifies two consequences of this choice: more uniform load sharing and avoidance of over-pushing any single leg.

The software stack is built in ROS 2 with Ignition Gazebo. The robot is modeled in URDF, including the body, three legs, drill unit, and force sensors. Joint control is provided through ros2_control, with prismatic joints operated in velocity control and revolute joints in position control. The main ROS 2 components are an FSM node, which subscribes to joint states and force-sensor topics and publishes joint commands and mission status; a low-level actuator node on the Arduino, which executes commands and reports encoder and limit-switch states; and logging and visualization nodes for force time series and simulated execution. The architecture is deliberately modular so that other controllers, including more advanced force control, can replace the threshold-based FSM.

5. Drilling integration and validation

In the Stinger Robot, drilling is not a downstream tool action appended to an anchoring system; it is structurally and algorithmically coupled to bracing. The drill module includes a revolute alignment joint, a prismatic feed joint, and a force sensor at the bit. During operation, the drill is aligned approximately along the central leg’s axis, advanced until rock contact is detected through a force threshold, and then rotated while feed continues as permitted by structural loading (Liu et al., 31 Jul 2025).

This integration matters because drilling imposes axial force along the drill axis, torque about that axis, and potentially lateral components if the drilling direction is misaligned or the rock is heterogeneous. The robot counters these through the central leg, which provides the primary reaction to axial drilling force, and through the triangular three-leg geometry, which resists torque and bending moments. The controller maintains hard-bracing preload during drilling, while the self-locking worm gear joints and screw-driven linear actuators prevent backdriving that would otherwise loosen the contact geometry.

The principal simulation experiment executes a full mission sequence: opening, initial bracing, hard bracing, and drilling. Force time series and a force-ratio measure show correct threshold-triggered state transitions, stable force distribution, and expected load transfer during drilling. In these simulations, the central leg increasingly absorbs axial load along the drill axis, validating the mechanical design rationale that the drilling axis should be aligned with the strongest available support path.

The hardware validation reported is preliminary but specific. In a self-support tension test, a wooden frame was used to emulate tunnel walls. The side legs were rotated to a Y shape, the linear actuators extended until all three legs contacted the frame, the actuators stopped after force contact was established, and the external support holding the robot was removed. The robot then supported its own weight via tension in the three legs, demonstrating that the distributed self-anchoring concept works in hardware.

A separate rotary-joint step-response test used an OptiTrack motion capture system for ground truth on one side leg’s angle. A commanded step input of 1.57 rad, approximately 90°, produced a 90° rotation in about 2 seconds, and the encoder estimate matched OptiTrack to 0.02 rad, approximately $R = 80{:}1$2. This confirms that the worm-gear rotary joint, motor, and encoder provide the positional fidelity required for the opening phase of the FSM. At the same time, the paper is explicit that the prototype drill hardware remains a dummy representation and that full rock drilling with fragmentation, dust, and field disturbances remains future work.

6. Limitations, future directions, and terminological context

The present system has a constrained operating envelope. Workspace analysis shows that it is effective only across a range of tunnel cross-sections set by $R = 80{:}1$3, $R = 80{:}1$4, and the side-leg angle limits. If the tunnel is too large, the legs are forced toward maximum extension and the body loses repositioning flexibility. If the tunnel is too small, body interference and insufficient leg articulation become limiting. The design also presumes sufficient friction at the contact pads, but the coefficient of friction can vary substantially in wet, muddy, fractured, or fragile rock, and the paper notes that smooth or highly fragile surfaces might limit achievable normal force without inducing surface damage (Liu et al., 31 Jul 2025).

The current implementation is also limited in autonomy scope. The finite-state machine is threshold-based and does not yet incorporate advanced force or impedance control, model-based optimization, or autonomous navigation and localization; those capabilities are delegated to other robots in the broader mining concept. This clarifies another potential misunderstanding: Stinger is not presented as a standalone mine-exploration robot but as a drilling anchor module in a multi-robot system.

The future work identified is correspondingly modular. Planned directions include real-world deployment in operational or abandoned mines, evaluation under real rock conditions and environmental disturbances, more advanced force control such as impedance control for leg deployment, adaptive planning of leg configurations from three-dimensional perception of tunnel geometry, and better estimation of friction and contact properties. The authors also envision integration with different drilling technologies, including percussive and rotary tools, and extension to tasks beyond drilling, such as bolting, reinforcement, or rock sampling.

In broader robotics terminology, the expression “stinger robot” has been used in other, conceptually distinct senses. One example is a scorpion-inspired terrestrial robot in which an active tail, treated as a “stinger,” is included as an action variable in a PPO-based controller for waypoint navigation; there the tail functions primarily for balance and posture, not for distributed force bracing or hard-rock drilling (Agrawal et al., 2020). Other hazardous-environment platforms, such as the tensegrity robot Tribar, have emphasized impact resistance, autonomous locomotion, and cliff survival rather than wall-braced drilling (III et al., 25 Jan 2025). At the microrobotic scale, Bee$R = 80{:}1$5 has been discussed as a compact flapping-wing platform into which one could imagine adding a “stinger” payload, again in a sense unrelated to underground anchoring (Yang et al., 2019). Against these usages, the Stinger Robot described here is distinctive in naming a self-bracing underground drilling architecture whose core technical problem is the autonomous creation and maintenance of a rigid, load-bearing contact structure in confined rock tunnels.