Source-linked AI summary
Scientific Exploration of Challenging Planetary Analog Environments with a Team of Legged Robots
Philip Arm, Gabriel Waibel, Jan Preisig, Turcan Tuna, Ruyi Zhou, Valentin Bickel, Gabriela Ligeza, Takahiro Miki, Florian Kehl, Hendrik Kolvenbach, Marco Hutter
TL;DR
Planetary exploration robots struggle with steep, loose, and unstructured terrain, while single-robot missions provide limited speed and skill diversity. The paper presents a team of complementary legged robots combining locomotion, mapping, target segmentation, scientific instruments, and task-level autonomy. Across three challenging analog deployments, the approach supported effective missions and access to planetary targets currently unreachable by wheeled systems.
Problem
Many scientifically and resource-relevant planetary targets remain difficult to access because wheeled robots are limited on steep slopes, loose soil, and unstructured terrain, while single robots offer limited exploration speed and skills.
Method
The paper develops a team of legged robots with complementary roles, redundant payloads, efficient locomotion, mapping, target segmentation, scientific instruments, and autonomous task execution.
Results
The approach was validated in three challenging analog environments and demonstrated investigation of lunar and Martian targets currently unreachable by wheeled rover systems.
Takeaways & Limitations
Teamed legged robots can support quick, effective scientific exploration and prospection in challenging planetary analog environments.
Abstract
from arXiv · showhide
The interest in exploring planetary bodies for scientific investigation and in-situ resource utilization is ever-rising. Yet, many sites of interest are inaccessible to state-of-the-art planetary exploration robots because of the robots' inability to traverse steep slopes, unstructured terrain, and loose soil. Additionally, current single-robot approaches only allow a limited exploration speed and a single set of skills. Here, we present a team of legged robots with complementary skills for exploration missions in challenging planetary analog environments. We equipped the robots with an efficient locomotion controller, a mapping pipeline for online and post-mission visualization, instance segmentation to highlight scientific targets, and scientific instruments for remote and in-situ investigation. Furthermore, we integrated a robotic arm on one of the robots to enable high-precision measurements. Legged robots can swiftly navigate representative terrains, such as granular slopes beyond 25 degrees, loose soil, and unstructured terrain, highlighting their advantages compared to wheeled rover systems. We successfully verified the approach in analog deployments at the BeyondGravity ExoMars rover testbed, in a quarry in Switzerland, and at the Space Resources Challenge in Luxembourg. Our results show that a team of legged robots with advanced locomotion, perception, and measurement skills, as well as task-level autonomy, can conduct successful, effective missions in a short time. Our approach enables the scientific exploration of planetary target sites that are currently out of human and robotic reach.
INTRODUCTION
Planetary exploration targets can exceed wheeled robots’ terrain capabilities, motivating a redundant team of dynamically walking robots with complementary scientific and operational skills. The proposed system combines specialized robots, autonomy, perception, and measurement capabilities, and was validated across three challenging analog environments.
- Motivation: Hard-to-reach lunar targets and uncertain surface properties make efficient traversal without sacrificing scientific and resource-prospection capabilities a priority.Relevant targets include pyroclastic vents, volcanic rilles, caves, irregular mare patches, and fresh impact craters.
- System approach: The team uses three legged robots with complementary roles, redundant payloads, and high-level task autonomy to cover more area and collect more diverse scientific data than single-robot approaches.Operators assign navigation, remote-measurement, and in-situ-measurement tasks, while robots execute them autonomously and can continue during unreliable communication or signal loss.
- Motivation: Wheeled rovers provide robustness on relatively flat terrain but reach limitations on steep slopes, loose granular terrain, and unstructured environments.Planetary examples include Spirit becoming immobilized in loose soil and Opportunity temporarily sticking in a dune.
- Validation: The team was validated in three challenging analog environments: the ExoMars locomotion test facility, a quarry site, and the Space Resources Challenge competition site.The deployments were used to report results and lessons learned from the teamed exploration approach.
- System approach: The approach integrates efficient locomotion, online and post-mission mapping, instance segmentation for scientific targets, and remote and close-up instruments, including robotic-arm-enabled measurements.The system advances legged robots beyond locomotion toward realistic analog-mission interaction and scientific investigation.
- Validation: The robots demonstrated the potential to investigate scientifically transformative lunar and Martian targets that wheeled rover systems currently cannot reach.The paper frames this capability as applying to both martian and lunar analog environments.
A Team of Legged Robots for Planetary Exploration
A team of three four-legged robots combines complementary exploration, mapping, perception, locomotion, and scientific measurement capabilities for challenging planetary analog environments. Across the SRC and quarry deployments, the robots traversed difficult terrain, mapped sites, identified targets, and collected scientific data while maintaining redundancy and operational flexibility.
- System architecture: The team comprises the Scout, Hybrid, and Scientist, each with dedicated roles, while shared payloads and capabilities provide redundancy for task reallocation after failures.Two operators send high-level navigation and measurement goals, while the robots return state, maps, images, and scientific data through mission control.
- SRC deployment: During the SRC, the team mapped 95% of the competition area, located seven of eight boulders, identified 18 potential REAs, and performed targeted CTX-FW, MIRA, and MICRO measurements.Three robots operated simultaneously for 20 min, with the Scout mapping while the Hybrid and Scientist collected scientific data.
- Locomotion: The robots climbed and descended 25° slopes without locomotion failures, reached 0.7 m/s, and negotiated loose hills and high bedrock steps.The 25° slope was the maximum available in the Beyond Gravity testbed, and the robot completed three ascent and descent tests on both ES-4 and bedrock.
- Locomotion: The arm-compatible controller preserved baseline robustness while reducing power consumption by 15% with a static arm on flat ground.The moving arm did not impede the Scientist during flat-terrain walking, and the controller supported heavy payloads including the robotic arm.
- Mapping and science: Online mesh maps supported operations, dense point clouds supported postmission visualization, and instance segmentation highlighted boulders even under difficult lighting.The mapping and perception pipeline helped operators identify and prioritize scientific targets during the SRC and quarry missions.
DISCUSSION
The deployments show that specialized legged robots can increase scientific data collection and preserve mission goals despite payload malfunctions. Remaining barriers include limited autonomy, communication constraints, payload scope, and space-qualification challenges.
- Specialized legged robots substantially increased scientific data collection during limited-time planetary analog missions, including instrument deployment every 3–5 min at the SRC.A dedicated scout robot reduced the time other robots spent on mapping and target identification, improving high-return instrument utilization.
- Six of seven quarry mission goals were fulfilled despite two malfunctioning payloads, illustrating the practical value of redundancy.
- The payload suite balanced remote and in-situ science but was constrained by budget and time, while XRF instruments performed better at identifying resource-enriched areas.The authors identify XRF or LIBS integration and sample return as expansions that could increase scientific output.
- A 5 s communication round-trip time affected operations, requiring reduced data sizes and lightweight products to lower packet-loss risk.Task-level autonomy limited operator interaction to one exchange per task, while further protocol and compression analysis remains open.
- Operators still had to prioritize and allocate tasks, making decision-making time-consuming and motivating automated scientific interpretation, task allocation, and robot-to-robot collaboration.
- Scaling to actual space missions remains constrained by unavailable space-grade processors and non-space-qualified LiDARs, although solid-state LiDARs may support future deployment.
MATERIALS AND METHODS
The system combines legged robots, scientific payloads, robotic-arm deployment, locomotion control, and mapping for planetary-analog exploration and measurement. Its perception and mapping stack supports localization, dense visualization, lightweight online meshes, and target characterization.
- Hardware and payloads: The team uses ANYmal-based robots, including Scout, Hybrid, and Scientist, with the Scientist carrying a 6-DoF DynaArm for instrument deployment.ANYmal has a 15 kg payload capacity and 90 min nominal continuous-walking operation time.
- Hardware and payloads: The Scientist deploys MICRO and other instruments with its arm, while MICRO multispectral images support more detailed petrographic assessment than CTX-FW analysis.The arm’s distance feedback, linear actuator, and autofocus routine help MICRO acquire sharp images despite imprecise placement.
- Legged locomotion: The locomotion policy is adapted for payload disturbances by incorporating randomized payload mass and arm position and velocity observations.The modified policy targets robust, efficient motion on robots carrying scientific payloads and a robotic arm.
- Localization, mapping, and perception: CompSLAM provides LiDAR-based localization, while separate dense mapping maintains a 30 mm voxel map and filters registered, outlier, and inter-robot points.The mapping system also projects RGB information onto a copy of the point cloud for colorized maps.
- Localization, mapping, and perception: The system supports both high-resolution post-mission meshes and lightweight online meshes made from downsampled, compressed point clouds.Online transmission uses a 150 mm voxel size within a 9 m radius, while post-mission processing reconstructs a higher-resolution mesh.
SUPPLEMENTARY RESULTS
A laboratory mock mission compared the locomotion energy use of the authors’ controller with a baseline controller. After subtracting 175 W of standby consumption, the authors’ controller averaged 403 W of locomotion power.
- Power Consumption Experiments: The experiment used two waypoints 3.8 m apart, with the robot walking back and forth and stopping 8 s at each waypoint.The Scientist robot carried a static arm during the laboratory mock mission.
- Power Consumption Experiments: 403 W was the robot’s average locomotion power consumption with the authors’ controller after subtracting 175 W of standby use.The residual includes actuator power and increased consumption or losses in computers, fans, and power electronics.
- Power Consumption Experiments: Standby consumption was 175 W with all computers and actuators running in standby mode.This value was subtracted from subsequent measurements to isolate additional locomotion power consumption.
SUPPLEMENTARY METHODS
The supplementary methods detail locomotion training for robots carrying arms and payloads, arm control for measurement, and modular behavior-tree autonomy across the team.
- Locomotion Policy Details: The locomotion policy incorporates arm disturbances, joint measurements, payload-mass randomization, and curriculum training to improve robustness with additional equipment.The arm is simulated with independent PD control while the locomotion policy controls the legs; payload mass is randomized and provided as privileged teacher information.
- Arm Control Implementation Details: The arm controller models a floating-base 6-DoF manipulator and combines state deviation, tool-pose tracking, input, and self-collision costs.During walking, the controller prioritizes a stable, low-center-of-gravity arm configuration; during measurements, it enables end-effector tracking.
- Arm Control Implementation Details: MIRA alignment uses a desired 0.7 m stand-off distance and iteratively applies PID-controlled end-effector twists until alignment error falls below 0.01 m.The alignment error is transformed into a desired instantaneous velocity and sent through the robotic-arm twist interface.
- Arm Control Implementation Details: MICRO deployment combines direct pose commands with a PID twist loop whose distance reference compresses the sensor’s soft foam against the target without damaging it.The selected reference compression distance is 0.005 m.
- Behavior Trees for Robust Task-Level Autonomy: Behavior trees provide modular, reusable navigation, inspection, and measurement behaviors that are minimally adapted to each robot’s payload configuration.The Scientist behavior tree contains 76 leaf nodes, while operators can monitor, cancel, or replace running actions and adjust parameters.