Autonomous Whole-body Loco-manipulation Using a Snake Robot

Adarsh Salagame, Henry Noyes, Haroon Hublikar, Alireza Ramezani

November 2026

TL;DR The 11-joint snake robot COBRA pushes a box to a goal with its whole body while sidewinding, steered by a Bézier planner and a proportional controller on a reduced-order pose. It delivered the box within 0.2 m of the goal in all 20 hardware trials.

What they do

A snake robot has no separate base and arm: the body that propels it is also what touches the object. Earlier snake loco-manipulation mostly used a gripper on one end, or ran whole-body loco-manipulation open loop with pre-defined gait sequences. Here COBRA pushes a box to a goal under closed-loop control while sidewinding, and any part of its body may make contact. The paper also builds a contact-implicit model and trajectory optimization for this task, but validates only a simplified version of the optimization, in simulation.

How it works

Contact-implicit model. Moreau-Jean time stepping with projected Gauss-Seidel resolves robot-ground and robot-box contacts within friction cones. The box is a planar rigid body with two ground contacts.
Reduced-order pose. The oscillating head makes a poor reference, so the robot is planned as its center of mass, a principal-axis heading and a bounding box.
Planner and controller. A quadratic Bézier path leads to the box, then a cubic one runs through the box to the goal. Pure pursuit and a proportional law set one steering bias on a sidewinding gait, which also seeds the trajectory optimization.
Robot and box poses come from motion capture.

Results

In 20 hardware trials with varied start, box and goal poses, the robot delivered the 0.5 kg box to within 0.2 m of the goal every time, sidewinding at about 0.1 m/s.
A longer run pushed the box through two waypoints, with heading swings over 200° while the steering bias stayed within ±15°.
In simulation, the reduced optimization finds pushing forces that reproduce a recorded box trajectory within friction limits. It solves the ~145 s run in under a minute.

Limitations

The hardware runs use the planner and nominal controller, not the full trajectory optimization, and use no contact feedback. The optimization was checked only on the box subsystem, in simulation. Tests ran on flat, rigid ground with motion capture. Onboard localization and object detection are left for future work.

Figures

From Figure 1: COBRA demonstrating closed-loop loco-manipulation, pushing an object toward the goal. The green footpr... — From Figure 1: COBRA demonstrating closed-loop loco-manipulation, pushing an object toward the goal. The green footprint is its reduced-order representation.

From Figure 2: Full-order and reduced-order representations of the snake robot. Left: full-order model with floating-... — From Figure 2: Full-order and reduced-order representations of the snake robot. Left: full-order model with floating-base pose, joint coordinates, robot-ground contact impulses with their friction cone, and the bounding-box projection. Right: reduced-order planning space with robot start pose (R), box start pose (B) and goal (G); the robot reference trajectory has an approach phase and a loco-manipulation phase.

From Figure 5: Hardware loco-manipulation results. Top left: composite snapshots of a trial with robot start (R), box... — From Figure 5: Hardware loco-manipulation results. Top left: composite snapshots of a trial with robot start (R), box start (B) and goal region (G). Top right: the matching plot of reduced-order bounding box, robot center-of-mass path, box path and planned Bézier path. Bottom: trajectories from four more trials with different start and goal configurations.

From Figure 7: Waypoint-tracking experiment and contact model check. Top: snapshots and trajectories as the robot pus... — From Figure 7: Waypoint-tracking experiment and contact model check. Top: snapshots and trajectories as the robot pushes the box through waypoints WP1 and WP2 to the goal. Bottom: robot-box contact forces against their friction cone, box-ground friction forces against the friction polyhedron, and the simulated box trajectory overlaid on the experimental one.

Abstract

Snake robots can leverage their hyper-redundant morphology for both locomotion and manipulation through distributed body contact, but whole-body loco-manipulation is particularly challenging since the entire body serves simultaneously as propulsive mechanism and manipulation tool. Prior work on wheeled, legged, and snake robot platforms has decoupled these functions through dedicated appendages, body segmentation, or grippers. We present closed-loop loco-manipulation with our snake robot COBRA without this distinction, such that any part of the body may make contact. We develop a contact-implicit dynamic model unifying robot–ground, robot–object, and object–ground interactions within a Moreau–Jean time-stepping framework, and formulate a trajectory optimization over the full-order dynamics, guided by a geometric planner and nominal proportional controller, to autonomously push an object to a goal location during sidewinding. We validate the planner and controller on hardware across varied configurations and present simulation results for a simplified trajectory optimization that resolves robot–object and object–ground contacts.

Summaries are written automatically from each paper's text. Figures and videos come from the paper and its project page; all credit goes to the authors.

Matt Strong