Source-linked AI summary
LIMBERO: A Limbed Climbing Exploration Robot Toward Traveling on Rocky Cliffs
Kentaro Uno, Masazumi Imai, Kazuki Takada, Teruhiro Kataonami, Yudai Matsuura, Antonin Ringeval-Meusnier, Keita Nagaoka, Mikio Eguchi, Ryo Nishibe, Kazuya Yoshida
TL;DR
Steep, uneven planetary terrain remains difficult for conventional wheeled robots, and prior climbing robots leave a need for improved spine gripping in rigorous conditions. LIMBERO combines a coupled-motion spine gripper with terrain graspability assessment. It climbed rocky terrain under 1 G, a demonstration not previously achieved by 10 kg-class limbed robots of this type.
Problem
Conventional wheeled robots struggle on rough, uneven terrain, while most climbing robots are compact and the robust larger-scale example cited was demonstrated under reduced gravity.
Source: §I · pp. 1–2Method
LIMBERO combines a single-motor gripper with coupled finger-closing and spine-hooking motions, a terrain graspability algorithm, and a quadrupedal climbing platform.
Source: §I · p. 2Results
Across ten trials, LIMBERO completed 44 steps and climbed rocky terrain under full Earth gravity, averaging 0.5 m ascent in 30 min.
Source: B. Rock Climbing Experiment · p. 7Takeaways & Limitations
The experiments demonstrate stable rocky-cliff hanging and climbing under 1 G, alongside terrain assessment for climbing tasks.
Source: §VI · p. 7
Abstract
from arXiv · showhide
In lunar and planetary exploration, legged robots have attracted significant attention as an alternative to conventional wheeled robots, which struggle to traverse rough and uneven terrain. To enable locomotion over highly irregular and steeply inclined surfaces, limbed climbing robots equipped with grippers on their feet have emerged as a promising solution. In this paper, we present LIMBERO, a 10 kg-class quadrupedal climbing robot that employs spine-type grippers for stable locomotion and climbing on rugged and steep terrain. We first introduce a novel gripper design featuring coupled finger-closing and spine-hooking motions, tightly actuated by a single motor, which achieves exceptional grasping performance (>150 N) despite its lightweight design (525 g). Furthermore, we develop an efficient algorithm to visualize a geometry-based graspability index on continuous rough terrain. Finally, we integrate these components into LIMBERO and demonstrate its ability to ascend steep rocky surfaces under a 1 G gravity condition, a performance not previously achieved yet for limbed climbing robots of this scale.
I. INTRODUCTION
Conventional wheeled robots have limited access to steep, uneven planetary terrain, while existing climbing robots leave a need for stronger spine gripping in more rigorous conditions. LIMBERO combines a coupled-motion gripper and terrain graspability algorithm, then demonstrates rocky-terrain climbing under full Earth gravity.
I. INTRODUCTION
- Wheeled robots offer efficiency on smooth sandy terrain, but steep and uneven sites such as lunar skylights and Martian outcrops remain difficult to access.Legged robots can select contact points and jump or climb over rough terrain –.Source: §I · p. 1
- Earlier climbers include LEMUR 3 and compact designs such as RockClimbo, LORIS, HubRobo, and SCALER, but most remain minimal and compact.LEMUR 3 demonstrated robust climbing while supporting relatively large mass, but under reduced gravity; other designs used numerous claws or reduced gripper size and degrees of freedom.Source: §I · pp. 1–2
- LIMBERO demonstrated stable rocky-terrain climbing under full Earth gravity while supporting its own weight and a 1.4 kg payload; the authors call it the first relatively large-scale demonstration of its kind.The robot is described as 10 kg-class.Source: §I · p. 2
- The gripper couples finger-closing and spine-hooking motions through one motor and a reduction gear, achieving grasping performance above 150 N.The design targets stronger engagement through both biting and hooking motions.Source: §I · p. 2
- The algorithm visualizes geometry-based graspability for local points across continuous rough terrain.It is intended to identify suitable graspable geometries from terrain.Source: §I · p. 2
II. SPINE GRIPPER DESIGN
LIMBERO’s lightweight gripper demonstrates strong engagement with rocky terrain, supporting a 15 kg load and a 10 Nm external moment. The surplus performance is intended to provide a safety margin against unexpected gripper detachment.
II. SPINE GRIPPER DESIGN
- A 0.525 kg gripper supported a 15 kg load and withstood a 10 Nm external moment while grasping pumice on a vertical wall.Fig. 2 presents the developed gripper’s grasping performance.Source: §II · p. 2, Figure 2 · p. 3
- The gripper’s excess performance is intentional, providing a safety margin if one or more grippers detach unexpectedly.Source: §II · p. 3
A. Principle
The gripper couples spine biting and hooking to improve engagement, using a single actuator and a compact linkage mechanism. Its cable-driven sequence closes the fingers before scratching the spines across the surface.
A. Principle
- Substantially enhancing spine-gripper performance requires coupling the spikes’ biting and hooking actions.Earlier approaches used only one principle, while LEMUR’s many-spine hooking design is bulky and the GOAT gripper’s biting approach struggles on smoother or gently rough terrain.Source: A. Principle · p. 3
B. Mechanism
- Eight radially arranged fingers carry 16 spines, and one actuator drives both biting and hooking motions through a linkage.The dual spines sit on passive roll joints for adaptability; Fig. 3 shows the motor, worm gearing, translating stage, and wires.Source: B. Mechanism · p. 3, Figure 3 · p. 3
- A cable pull closes the fingers for biting before scratching the spines for hooking; releasing the cable reverses the sequence.A roughly 25:1 worm reduction tensions the cables, and its mechanical lock holds position without continuous actuator current.Source: B. Mechanism · p. 3
- The flexible wire loop balances finger loads to some extent, while individual springs help the fingers adapt to arbitrary terrain shapes.The design is more compact and simpler than the load-sharing mechanism in, but does not match its load-sharing performance.Source: B. Mechanism · p. 3
III. LIMBED CLIMBING ROBOT DESIGN
LIMBERO is a 9.7 kg quadrupedal climbing robot built around four limbs and a central base. Its parallel-linked limbs reduce inertia while retaining workspace and provide active positioning and passive yaw at the gripper.
III. LIMBED CLIMBING ROBOT DESIGN
- LIMBERO’s total mass is 9.7 kg, comprising a 2.8 kg central body and four limbs weighing 1.72 kg each, including grippers.The prototype uses CFRP tubes and 3D-printed components, with stainless steel and aluminum reserved for parts needing high durability.Source: §III · pp. 3–4
A. Limb Design
- Each limb has four active degrees of freedom and one passive yaw degree of freedom at the gripper, with a parallel linkage to reduce inertia while preserving workspace.The four active joints provide three translational end-effector degrees of freedom and a pitch joint that keeps the gripper palm parallel to terrain; Fig. 4 shows the parallel-linked design and 3.5:1 pitch-joint reducers.Source: A. Limb Design · p. 4, Figure 4 · p. 4
- The passive gripper yaw joint prevents torsional moments at the leg tips when the base moves while all four grippers remain engaged.Source: A. Limb Design · p. 4
B. Base Design
The base houses LIMBERO’s electronics, sensors, and batteries in an axisymmetric structure. Its software links perception, planning, and control, while graspability assessment focuses on terrain geometry suitable for spine hooking.
B. Base Design
- The axisymmetric base houses onboard computing, an IMU, motor controllers, and a 3D LiDAR, with batteries at the bottom to lower the center of gravity.It uses 3D-printed plastic and CFRP pipes; Fig. 5 depicts the base structure and its accommodation of electronics, sensors, and batteries.Source: B. Base Design · p. 4, Figure 5 · p. 4
C. Software Design
- LIMBERO’s software has perception, planner, and controller modules, with limb commands passing through high- and low-level control to joint controllers.State estimation supplies robot pose and end-effector contact states to the planner; Fig. 6 shows the software architecture.Source: C. Software Design · p. 4, Figure 6 · p. 5
IV. GRASPABILITY ASSESSMENT
- Graspability assessment detects geometrically graspable shapes in a 3D terrain map, assuming rough rocky surfaces with micro-asperities suitable for spine hooking.Shape and roughness can be sensed or predicted from terrain data, whereas stiffness and fragility require contact-based assessment.Source: §IV · p. 5
A. Problem Formulation
Geometric graspability assessment identifies terrain points whose local shape fits the gripper’s graspable geometry. It takes a raw point cloud, gripper parameters, and voxel-grid size, then scores and filters terrain points.
A. Problem Formulation
- The algorithm takes a raw point cloud P, gripper parameters Γ, and voxel-grid size c as inputs.P contains n spatial points; Γ defines the gripper’s gripping range of motion.Source: A. Problem Formulation · p. 5
- It constructs terrain and gripper arrays, extracts a terrain subset around each point, and scores its match to the gripper mask.The algorithm returns graspability scores g and the set G of points exceeding a predetermined threshold.Source: A. Problem Formulation · p. 5
- The score represents each point’s suitability for grasping, with points above threshold ˆg included in G.The score is computed for points in the preprocessed terrain array.Source: A. Problem Formulation · p. 5
B. Terrain Array
The terrain array represents the ground surface as a three-dimensional Boolean voxel grid. It is built by transforming and homogenizing point-cloud data before voxelization.
B. Terrain Array
- The terrain array is an efficient representation of the three-dimensional ground-surface shape.Source: B. Terrain Array · p. 5
- Point-cloud coordinates are transformed so the z-axis aligns with the normal of a regression plane fitted to the cloud.Source: B. Terrain Array · p. 5
- Occluded areas are linearly interpolated by Delaunay triangulation to homogenize point-cloud density before voxelization.Source: B. Terrain Array · p. 5
- Voxelization labels each cell solid (1) or void (0), yielding a three-dimensional Boolean array.The terrain-array dimensions are determined by the voxel grid size.Source: B. Terrain Array · pp. 5–6
C. Gripper Mask
The gripper mask encodes the volume that the gripper can grasp, while geometric graspability assessment compares that mask with local terrain geometry. The resulting scores distinguish convex summits from locally sloped or concave regions.
C. Gripper Mask
- The gripper mask is a three-dimensional array representing graspable volume, defined by the gripper’s design parameters Γ.For mechanical fingered grippers, its shape can reflect the fingers’ angular range of motion.Source: C. Gripper Mask · p. 6
D. Geometric Graspability Assessment
- At each terrain point, the algorithm compares a same-sized local terrain subset T* with mask M to compute a graspability score.The score is based on the matching solid voxels relative to the number of solid voxels in T*.Source: D. Geometric Graspability Assessment · p. 6
- Higher voxelization resolution produces more accurate graspability scoring but requires longer computation time.The output is an interpolated point cloud whose colors represent graspability scores.Source: Figure 8 · p. 6
- A z-threshold can reduce computational cost by restricting evaluation to points above the threshold.Source: D. Geometric Graspability Assessment · p. 6
- In the Fig. 7 case study, graspability scores are examined for different terrain shapes using two-dimensional voxel arrays.Source: D. Geometric Graspability Assessment · p. 6
E. Result
The graspability algorithm scored a pumice terrain scan, assigning higher scores to convex summits and lower scores to locally sloped or concave regions. LIMBERO also demonstrated a payload capacity of approximately 1.4 kg under full Earth gravity in a static cliff-hanging test.
E. Result
- With 108,745 input points and 2 mm voxels, graspability assessment scanned the full terrain in 64 s on LIMBERO’s Jetson Orin NX.No z-threshold was used, so concave regions were included.Source: E. Result · p. 6
- On randomly distributed pumice, convex summits received higher scores, while locally sloped or concave regions received lower scores.The authors interpret this pattern as effective evaluation of geometrical graspability across the terrain map.Source: E. Result · p. 6
V. EXPERIMENTS
- For climbing tests, LIMBERO was teleoperated on a 60° inclined field after LiDAR prescanning and offline graspability evaluation.The field had randomly placed pumice stones, and the operator selected regions with high graspability scores.Source: §V · p. 6
A. Rocky Cliff Holding Experiment
- Approximately 1.4 kg payload capacity was indicated under full Earth gravity in a static cliff-hanging test.LIMBERO stably supported its body and an additional 2.5 kg load; the payload estimate accounts for omitted battery mass.Source: A. Rocky Cliff Holding Experiment · p. 7, Figure 9 · p. 7
B. Rock Climbing Experiment
LIMBERO climbed rocky terrain under full Earth gravity in teleoperated trials, while exposing joint-torque, gripper-slip, and autonomy challenges that motivate further development.
B. Rock Climbing Experiment
- A human operator selected swing limbs, set limb reference velocities, and issued gripping and releasing commands using a preobtained graspability-assessed terrain map.The foothold sequence was manually planned from that map.Source: B. Rock Climbing Experiment · p. 7
- Across ten trials, LIMBERO completed 44 steps, averaging 3.8 min per step and 0.017 m/s while ascending 0.5 m in 30 min under full Earth gravity.The reported climbing speed was slowed by manual gripper-position adjustments; actuator power averaged 10 W, ranging from 5–30 W.Source: B. Rock Climbing Experiment · p. 7
- Shoulder and wrist joints sometimes exceeded torque limits, requiring reboot commands and occasionally halting tests for safety; supporting grippers also slipped during upward motion, with one unintended detachment in ten trials.Despite intermittent slipping, the grippers usually maintained their grasp.Source: B. Rock Climbing Experiment · p. 7
C. Lessons Learned
- Moving the base upward required significant joint torque, and strongly adhered limbs can concentrate internal forces on particular joints, motivating stronger shoulders and flexibility or force-accommodating control.The authors identify reaction-mitigating motion planning and impedance control as possible strategies.Source: C. Lessons Learned · p. 7
- Spine tips showed slight wear after more than 100 grasping cycles, although the study observed no noticeable performance decline on its lava rocks.The spikes were vacuum-hardened for high durability.Source: C. Lessons Learned · p. 7
- Terrain-geometry graspability scores cannot account for unexpected factors such as surface breakage, so practical deployment also requires autonomy to assess whether candidate contacts are slippery, broken, or too soft.The proposed workflow includes scanning and mapping terrain, detecting graspable geometry, then haptically assessing candidate contacts.Source: C. Lessons Learned · p. 7
VI. CONCLUSION
The conclusion presents LIMBERO’s coupled-motion spine gripper, terrain graspability algorithm, and full-gravity rocky climbing demonstration. It identifies limb-joint improvements and more autonomous, torque-aware control as future needs.
VI. CONCLUSION
- LIMBERO combined a coupled biting-and-hooking active spine gripper with an efficient algorithm for extracting graspable geometry from 3D terrain maps, and demonstrated rocky cliff climbing under 1 G.The authors report stable cliff-hanging and climbing in real-world experiments.Source: §VI · p. 7
- The experiments indicate that limb joints need further enhancement for upward movement on vertical surfaces, a requirement the authors identify for planetary and terrestrial deployment.The outcomes are presented as a foundation for more capable and autonomous limbed climbing robots.Source: §VI · p. 7
- Future work includes autonomous foothold selection among discrete graspable points and control that reduces contact reaction forces and balances joint torques,, while maintaining tumble-stability criteria,.The authors also call for further hardware sophistication.Source: §VI · pp. 7–8