Mapless, Stabilizing, and Anti-Stuck Flipper Control for Articulated Tracked Robots

Valentýn Číhala; Pavel Sarga; Martin Pecka; Tomáš Svoboda; Karel Zimmermann

cs.ROOct 9, 2026

Mapless, Stabilizing, and Anti-Stuck Flipper Control for Articulated Tracked Robots

Authors: Valentýn Číhala, Pavel Sarga, Martin Pecka, Tomáš Svoboda, Karel Zimmermann

Organizations: Czech Technical University in Prague, Faculty of Electrical Engineering, Department of Cybernetics

Abstract

Nearly every autonomous or semi-autonomous flipper controller assumes access to an elevation map of the terrain under and in front of the robot. The terrain beneath the chassis is occluded in the sensors, so the map is remembered through odometry integration, whose drift corrupts it. We present Stabilizing Anti-Stuck (SAS), a mapless controller that combines a stabilization term for body roll and pitch with an anti-stuck term that frees the robot when it gets stuck. As input, SAS uses the IMU, the flipper joint angles, and a motion cue that only tells whether the robot is moving. The cue is computed over a short window and never accumulated, so SAS does not inherit the odometry-integration drift that corrupts the map. We deploy SAS on a tracked robot with four flippers and evaluate it against manual teleoperation and prior flipper controllers in a large comparison, both in simulation and on a real obstacle. In simulation and on the real robot, SAS achieves traversal quality comparable to state-of-the-art map-based controllers. On a log hidden in tall grass, invisible to the LiDAR, the best map-based controller gets stuck while SAS traverses the hidden obstacle without a problem.

Figures & tables

Panel of Figure 1 — Fig. 1 : The onboard and ground-truth map on a pallet. (a) The robot climbing a pallet obstacle. (b) The same pose, rendered from the data the robot observed there: the onboard map (blue) against the handcrafted ground-truth map positioned by AprilTags (transparent red), with the ground-truth robot pose drawn transparent. The offset between the two surfaces, and between the two bodies, is the map and localization error that the perception stack introduces. The spike behind the robot in the onboard map is a reflected, invalid LiDAR return.

Controller	Approach	Code	Map	Stab.	Anti- stuck
Manual
Manual control [ 12 ]	direct teleoperation	–	$\times$	–	$\times$
Manual modes [ 8 ]	operator switches preset modes	–	$\times$	–	$\times$
Reactive and geometry-based
Číhala GC/OA [ 12 ]	clearance + obstacle-aware	released	✓	–	$\times$
Chen GFMP [ 2 ]	pose prediction + dynamic prog.	reimpl.	✓	p	$\times$

TABLE I : Compared controllers. Map : needs an onboard map. Stab. : attitude axes with an explicit stabilization objective (r roll, p pitch). Anti-stuck : frees itself when stuck. Code : the authors’ released implementation or our reimplementation.

Fig. 2 : Simulation arena. The warehouse course: a straight aisle of twelve numbered obstacles of variable difficulty (pallets, stacked boxes, planks, and a staircase); the robot starts at the right. In the autonomous run (Sec. IV ) it is driven at a constant $0.6\text{\,}\mathrm{m}\text{${\cdot}$}{\mathrm{s}}^{-1}$ through every obstacle in turn, with a $1\text{\,}\mathrm{m}$ settling strip between obstacles. — Fig. 2 : Simulation arena. The warehouse course: a straight aisle of twelve numbered obstacles of variable difficulty (pallets, stacked boxes, planks, and a staircase); the robot starts at the right. In the autonomous run (Sec. IV ) it is driven at a constant $0.6\text{\,}\mathrm{m}\text{$ {\cdot} $}{\mathrm{s}}^{-1}$ through every obstacle in turn, with a $1\text{\,}\mathrm{m}$ settling strip between obstacles.

Channel pair	Simulation	Real
Shock – clearance	$0.26$	$0.06$
Shock – tilt	$0.28$	$0.39$
Clearance – tilt	$0.39$	$-0.03$

TABLE II : Correlation of the channel scores over the traversed simulated obstacles and real crossings.

	Simulation, twelve obstacles					Real pallet, one ride of ten crossings
Controller	TQ $\uparrow$	Shock $\uparrow$	Clearance $\uparrow$	Tilt $\uparrow$	Override [%] $\downarrow$	TQ $\uparrow$	Shock $\uparrow$	Clearance $\uparrow$	Tilt $\uparrow$	Override [%] $\downarrow$
	each cell: ground-truth map / onboard map
Manual
Manual control [ 12 ]	– / 0.888	– / 0.922	– / 0.866	– / 0.875	–	– / 0.810	– / 0.775	– / 0.714	– / 0.942	–
Manual modes [ 8 ]	– / 0.846	– / 0.916	– / 0.788	– / 0.834	–	– / 0.810	– / 0.766	– / 0.816	– / 0.848	–
Reactive and geometry-based

TABLE III : Traversal quality in simulation and on the real pallet. Simulation: mean over ten runs of the twelve-obstacle course. Real: one ride of ten crossings per controller and map, mean over the crossings. All channels are scored against ground truth, with

s_{\max}

376\text{\,}\mathrm{m}\text{

{\cdot}

}{\mathrm{s}}^{-2}

in simulation and

57.5\text{\,}\mathrm{m}\text{

{\cdot}

}{\mathrm{s}}^{-2}

on the real robot. Each cell gives the value under the ground-truth map / the onboard map. Shock, clearance and tilt are the three channel scores of ( 7 ). Override : percentage of distance an anti-stuck dig drove the flippers, by the SAS override for the baselines and by its own term for SAS. Manual rides are hand-driven from a third-person view. Per column and map condition, the best controllers are in bold and the second best underlined (ties share the mark).

Fig. 3 : Traversal of an obstacle hidden in tall grass by the map-based controller Yuan [ 1 ] and by SAS. (a) The onboard map built from the LiDAR point cloud during Yuan’s crossing: the LiDAR does not penetrate the grass, and so the map shows only the canopy while the log remains obscured. (b) Yuan interprets the gap created by the log in the canopy as a drop and lowers the front flippers to descend; the lowered flippers slide under the log, causing the robot to get stuck underneath it. (c) SAS does not rely on the LiDAR-derived map; therefore, it keeps its default posture and easily crosses the log.

Fig. 4 : Onboard-map error over an obstacle in simulation and on the real pallet. Median over all rides against the progress through the obstacle. Top: the height difference between the onboard and the ground-truth map over the body footprint. Bottom: the share of cells off by more than $10\text{\,}\mathrm{cm}$ . Both grow while the robot ascends the obstacle and fall again on the descent.

Explore similar work

Language-Guided Terrain-Adaptive Neural MPC for Autonomous Traversal of Articulated Tracked Robots
Sep 11, 2026Zhenfeng Gan, Yanbo Chen, Lirong Che +3Constrained Motion Planning Language Model-Based Control
A Reconfigurable Tracked Robot for Enhanced Obstacle Traversal Through Movable Articulation Point and Internal Mass Relocation
Aug 7, 2026Yuki Uda, Yasutaka Nakashima, Motoji Yamamoto +1Robotics Terrain-Aware Robot Locomotion
Department of Mechanical Engineering, Kyushu University, 744 Motooka, Nishi-ku, Fukuoka 819-0395, Japan · Graduate School of Informatics and Engineering, The University of Electro-Communications, 1-5-1 Chofugaoka, Chofu-shi, Tokyo, 182-8585, Japan
Safe Local Navigation for Ackermann-Steered Robots in Unmapped Environments
Jun 18, 2026Christian Schaible, Shahin SirouspourRobot Navigation Safe Robot Navigation
Department of Electrical and Computer Engineering, McMaster University, Hamilton, ON L8S 4L8, Canada

Sep 11, 2026, cs.ROJ/K move · Enter open · S save

Language-Guided Terrain-Adaptive Neural MPC for Autonomous Traversal of Articulated Tracked Robots

Zhenfeng Gan, Yanbo Chen, Lirong Che, Yongyi Ma+2

Constrained Motion Planning Language Model-Based Control Terrain-Aware Robot Locomotion Model Predictive Control Search-and-Rescue Robotics

Open paper AskPDF arXiv

In urban search and rescue, articulated tracked robots (ATRs) must traverse structured but contact-rich environments such as stairwells and cluttered building interiors. Reliable autonomy remains challenging because robot-terrain interaction (RTI) is hybrid and discontinuous, and effective flipper-track coordination is difficult to model analytically. We present ASTRIL-MPC, a language-guided neural kinematics model predictive control (MPC) framework for autonomous traversal. A learned kinematics model predicts short-horizon task-state increments from a height sequence and recent trajectories; NMPC plans with multi-objective costs and strict feasibility constraints; and a large language model (LLM) proposes bounded updates to selected weights and bounds through a safety-checked interface with range clipping, rate limiting, and consistency checks. The compiled predictor enables a full control cycle within 100 ms. Across three traversal tasks and a multi-height generalization setting, ASTRIL-MPC improves an aggregate traversal-quality score by up to 71% over a non-adaptive NMPC and by 67% over a PPO baseline, while eliminating measurable collision impacts during descent. These results indicate that combining terrain-conditioned neural kinematics, optimization-based planning, and language-guided adaptation yields data-efficient and robust autonomy for articulated tracked robots. Real-robot trials over four indoor obstacles further demonstrate transfer to contact-rich physical traversal.

Abstract

Explore similar work

Language-Guided Terrain-Adaptive Neural MPC for Autonomous Traversal of Articulated Tracked Robots

A Reconfigurable Tracked Robot for Enhanced Obstacle Traversal Through Movable Articulation Point and Internal Mass Relocation

Safe Local Navigation for Ackermann-Steered Robots in Unmapped Environments