Source-linked AI summary
GeoTrussRover: Morphological Computation with Contact-Semantic Control Primitives
Muyuan Ma, Yi Zhang, Yang Yang, Xuanyan Zheng, Ruiqi Hu, Boxuan Ke, Zhenyu Chen, Yicong Lin, Xin Hao Yang, Daliang Xiao, Zhinan Hou, Wanhao Niu, Yuan Sun, Yan Yang, Yue Xie
TL;DR
GeoTrussRover addresses the high-dimensional coordination problem of using a load-bearing variable-geometry truss with a mobile base to traverse steps. It extracts contact-semantic primitives from one 21-member traversal, projects them onto feasible target-height configurations, and tracks them with a full-space QP. Transfer from 0.10 m to 0.075 m reduces objective-function evaluations by 63.7%, while feasibility analysis spans 0.10–0.46 m and the electric prototype traverses 2.11 wheel radii.
Problem
A variable-geometry truss can change contact geometry for obstacle traversal, but coordinating its coupled members with a mobile base creates a high-dimensional planning problem.
Method
The method extracts four contact-semantic primitives from a constrained 21-member traversal, projects them onto a target-height feasible set, and tracks the adapted motion with a full-space QP.
Results
63.7% fewer objective-function evaluations are required for transfer from 0.10 m to 0.075 m than for full recomputation.
Takeaways & Limitations
The load-bearing morphology stores reusable task coordination while full-space feedback preserves local correction during locomotion.
Takeaways & Limitations
The sim-to-real dynamic coupling evidence uses uncalibrated contact separation and exceeds the member-speed limit, so it supports a coupling explanation rather than a validated controller result.
Abstract
from arXiv · showhide
Reconfigurable robots can change their contact geometry when a fixed body cannot negotiate an obstacle. A variable-geometry truss (VGT) distributes this shape change through a load-bearing structure, but coupling it to a mobile base creates a high-dimensional coordination problem. GeoTrussRover combines an electrically actuated VGT, a wheeled base, and contact-semantic morphology planning and control. We solve one source traversal and extract four contact-semantic primitives that describe coordination among 21 members. Physics-constrained projection adapts them to unseen step heights with the same contact topology. When every phase remains feasible, adaptation does not recompute the complete motion. If one phase violates the new physical constraints, only that phase is recomputed. A full-space QP then tracks the adapted motion and corrects member and wheel errors. For transfer from 0.10m to 0.075m, the method reduces objective-function evaluations by 63.7% relative to full recomputation. Contact-phase feasibility analysis covers step heights from 0.10 to 0.46m, or 1.08 to 4.97 wheel radii, with the upper value near the theoretical feasible boundary. The electric prototype traverses 2.11 wheel radii. The resulting low-dimensional representation stores task coordination in a hyper-redundant, load-bearing morphology and reuses it during locomotion.
1 Introduction
GeoTrussRover addresses the coordination challenges of step traversal by distributing shape change through a load-bearing truss and reusing contact-semantic motion primitives. Its layered planner and controller adapt morphology to new step heights while preserving full-space correction.
- Motivation: A variable-geometry truss changes wheel placement, chassis height, support region, and load distribution, but coordinating its coupled members becomes costly when environmental constraints change.The truss replaces serial multi-axis limbs with distributed structural deformation, creating a high-dimensional constrained coordination problem.
- Motivation: Shared contact sequences across step heights allow one solved traversal to serve as a local coordinate chart for geometrically different tasks.Front engagement, edge clearance, forward support transfer, rear recovery, and morphology restoration retain the same ordering while member lengths vary.
- Approach: A primitive-guided full-space QP combines coordinated nominal morphology with full-dimensional corrections for pose, support, and contact errors.The controller preserves local correction freedom after primitive-based adaptation.
- Approach: The method extracts four primitives from a constrained 21-member source traversal and projects them onto a new contact-conditioned feasible set.Projection updates essential support configurations without recomputing the complete traversal when the target phases remain feasible.
- Contribution: The electrically actuated, load-bearing wheeled VGT is validated through theory, simulation, and physical experiments.The contribution is an electromechanical realization of distributed morphology-based locomotion.
2 Related Work
Prior work modifies contact geometry through limbs, wheels, mobile bases, modular structures, or embedded morphology, while planners and controllers coordinate these changes under physical constraints. GeoTrussRover differs by adapting reusable task-conditioned coordination on a feasible morphology set rather than relying on direct scaling or complete replanning.
- Morphological reconfiguration: Existing reconfigurable robots alter contact radius, contact mode, width, length, or wheel placement to improve terrain interaction and access.Related systems include transformable wheel-leg mechanisms, reconfigurable mobile bases, modular robots, tensegrity structures, and embedded shape-morphing modules.
- Planning and control: Whole-body planners and controllers coordinate pose, morphology, contact forces, and gait modes using optimization, MPC, learning, or supervisory state machines.These approaches provide physically meaningful solutions for specified environments while handling dynamics and inequality constraints.
- Movement primitives: Movement primitives enable low-dimensional reuse, but direct spatial scaling cannot ensure nonlinear truss closure, actuator travel, collision clearance, and contact-dependent load constraints.A truss primitive therefore must be adapted on a feasible morphology set rather than scaled geometrically.
- Morphological computation: Morphological computation embeds coordination in body geometry, material dynamics, or hardware connectivity, whereas task-conditioned representations can retain independently actuated structures.This distinction motivates storing reusable coordination in a distributed morphology without requiring fixed hardware coupling.
3 Method
GeoTrussRover combines a 21-member electrically actuated truss, four-wheel base, contact-conditioned morphology planning, and full-space feedback control. Four source traversal primitives are projected onto target-height feasible sets, then tracked while preserving actuator-level correction.
- Platform: The platform uses 21 independently actuated telescopic members in a load-bearing truss and four driven wheel modules for propulsion and steering.The two octahedral cells and triangular interface form a fixed nine-node graph; coordinated extension changes height, wheelbase, lateral span, and attitude.
- Architecture: The method separates graph closure, support feasibility, morphology reuse, and execution feedback across four interfaces.One source solve captures member coordination, target projection adjusts height-dependent geometry, and full-dimensional feedback corrects execution errors.
- Graph Kinematics: The closed truss is modeled by stacking member-distance constraints into graph closure, with node positions q and 21-member lengths ℓ.At regular configurations, the rigidity matrix has rank 21 and its six-dimensional null space contains rigid-body motions.
- Load Framework: Quasi-static feasibility combines member forces, active wheel–terrain forces, friction-cone constraints, and gravity scaling through the load multiplier β⋆.Each contact phase defines a local physically usable branch selected by equality constraints and inequalities.
- Primitive Coordinates: A constrained source traversal specifies four phase outcomes and extracts coordinated 21-member differences as contact-semantic primitives.The primitives parameterize the source path, while continuation knots provide smooth within-phase references.
- Adaptation and Control: Physics-constrained projection adapts each primitive to a target height by recovering closure, stroke, clearance, and support without recomputing the complete traversal.The projected reference is tracked by a 50 Hz whole-body QP using all member rates and wheel speeds, allowing correction of pose, load, and contact errors.
P1 P2 P3 P4 R (b) Member-stroke reserve over height and progress
Figure 5 characterizes how feasible 21-member morphologies vary with step height and semantic progress, while tracking remaining member stroke.
- Feasible configurations form height-conditioned morphology families across the first three principal coordinates, with colors distinguishing P1–P4 and restoration.
- Minimum remaining stroke is evaluated across all members as a function of step height and semantic progress.
4 Experiments and Results
The experiments evaluate contact-augmented actuation, primitive-guided whole-body control, transfer across step heights, and simulation-to-hardware execution. Primitive reuse reduces replanning while preserving feasible traversal and control corrections.
- 4.1 Why Morphology and Contact Matter: Contact-augmented actuation raises task authority at wall support, achieving about 1.5 times wheel-drive authority at front contact and 1.8 times at rear contact.It exceeds shape-only authority by more than 15 times at both phases.
- 4.1 Why Morphology and Contact Matter: Along the executed 0.10 and 0.075 m paths, α⋆ stays above 0.32 and β⋆ above 1.31, with rear-left lift forming the main support-transfer bottleneck.The denser height sweep finds larger corrections and lower stroke reserve during elevation and rear recovery.
- 4.2 Primitive Guidance in Full-Space Whole-Body Execution: Only PG-QP completes all three source, projected, and perturbed conditions, solving in about 4.17 ms versus 16.21 ms for RF-QP.PG-QP retains local correction directions while using primitive coordination.
- 4.3 Single-Source Transfer to Unseen Step Heights: 63.7% fewer objective evaluations are required for transfer from 0.10 m to 0.075 m, while both methods produce complete references and Primitive completes PG-QP traversal.Primitive uses 1588 evaluations versus 4374 for Full.
- 4.3 Single-Source Transfer to Unseen Step Heights: For transfer from 0.20 m to 0.30 m, phase-specific recomputation yields a successful route with 32.5% fewer evaluations than Full.Primitive uses 13,688 evaluations versus 20,271 for Full while both routes complete under the enhanced 250 N condition.
- Physical implementation: Isaac Sim completes the semantic sequence on a 0.46 m step, while the electric prototype traverses a 0.195 m step through dynamic shape change.
5 Discussion
The discussion frames morphology as an actuated part of control whose geometry and load-path changes complement wheel contact. It also identifies contact coupling and calibration limits that constrain interpretation of the dynamic diagnostic.
- 5.1 Morphology-Facilitated Motion Control: Combining shape actuation with wheel contact raises task-achievement ratios from 0.53 to 0.82 at front contact and from 0.41 to 0.74 at rear contact.The resulting coordination is stored for projection and PG-QP, reducing endpoint computation by 63.7%.
- 5.2 Dynamic Base and Shape Control Coupling in the Sim-to-Real Gap: Wheelbase, axle orientation, mass distribution, and wheel load remain coupled through contact during whole-body shape control.The diagnostic supports this coupling explanation rather than a validated controller result.
- 5.2 Dynamic Base and Shape Control Coupling in the Sim-to-Real Gap: The dynamic diagnostic uses uncalibrated contact separation and exceeds the member-speed limit, motivating actuator identification, calibrated collision geometry, and load-aware whole-body control.
6 Conclusion
GeoTrussRover reuses contact-semantic primitives from one solved traversal to adapt whole-body coordination across step heights while retaining local feedback. The method reduces planning effort, supports a broad feasibility range, and demonstrates traversal with an electric prototype.
- 63.7% fewer endpoint evaluations are achieved when transferring primitives from 0.10 m to 0.075 m.From 0.20 m to 0.30 m, locally recomputing rear recovery reduces evaluations by 32.5%.
- The feasibility analysis spans 1.08 to 4.97 wheel radii, with shape actuation and wheel contact acting complementarily.
- An electric prototype traverses 2.11 wheel radii through whole-body shape change.
- The load-bearing, hyper-redundant morphology stores reusable task coordination while preserving local feedback during locomotion.