Stanford Robotics Seminar ENGR319 | Autumn 2025 | Make Every Step an Experiment

The Role of Robots in Planetary Exploration [00:10]

• Robot-aided exploration is crucial for planetary missions, providing invaluable data and scientific discoveries that would otherwise be out of reach.
• As humanity prepares to return to the Moon and land on Mars, expectations for robot-aided missions are higher than ever, demanding robots that can safely reach destinations, operate reliably across diverse surfaces, and adapt to changing environmental conditions.
• Planetary surfaces present significant challenges, including large rocks, steep craters, and unconsolidated regolith that can cause robots to slip, sink, or become stranded, leading to costly mission setbacks.

"And that is certainly not a easy task because planetary surfaces are among the worst and most complicated environments that our robots must traverse and operate in."

Sensing Environmental Properties Through Interaction [03:53]

• A primary challenge in exploring unknown environments is gathering information about the terrain as the robot moves to make informed decisions.
• The research focuses on regolith mechanics and properties, which are critical for mobility, navigation, drilling, and construction, but difficult to discern by vision alone due to variations in compaction and appearance.
• Unlike vision, human and animal feet provide rich, dense information about surface properties (softness, compactness, wetness) with every step, which robots can similarly leverage.

"So for us humans and many of the animals when we are walking our feet is providing information that sometimes can be more informative than vision."

Proprioception-Based Sensing with High-Force Transparency Actuators [06:37]

• The approach avoids extra sensors by utilizing proprioception-based sensing, leveraging the feedback from the robot's own movements and interactions.
• Robots with legs made from actuators possessing high force transparency can be highly sensitive to external forces, enabling them to "feel" the world through interactions.
• Recent advancements in direct drive motors with improved torque density now make it feasible to use them in legged robots, allowing legs to serve as both locomotive limbs and sensors.

"So because of this high force transparency, if we make robot legs using these actuators, you can imagine that the leg can be super sensitive to any external forces the leg felt and that can build us robots that can sensibly feel the world from every single interaction."

Developing and Testing a Single Leg Sensor Unit [08:17]

• To characterize regolith mechanics, a single leg unit, named "Traveler," was developed using direct drive actuators with high force transparency.
• This leg is capable of various sensing protocols, such as poking and scraping, and can measure forces applied to the toe by sensing motor current and torque.
• The leg was mounted on a static test stand for geotechnical and planetary tests in outdoor environments and on other robots to decouple locomotion and sensing for data gathering.

"So John made a couple of different configuration of the leg for us to start our lab and field testing and validation."

Characterizing Regolith Strength with Force Sensing [10:28]

• The leg unit performs sensing protocols, such as vertical penetration and horizontal scraping, to measure normal and shear forces.
• These force measurements reveal rich signals about material strength, with the rate of force increase during penetration indicating stiffness and scraping measuring shear strength.
• Lab experiments using a fluidized bed, where sand strength is controlled by airflow, validated the leg's ability to accurately measure surface properties and stiffness.

"What you can see is during the green shaded region as like vertically penetrate into the soil the force FY here which is the vertical normal force increase almost linearly and rapidly with the penetration depths and that's and the rate of that increase is going to determine how soft the stiffness of the soil."

Field Testing in Analog Environments: Mount Hood and Yens [14:15]

• After lab validation, the leg unit was deployed in two field sites: a dune field in New Mexico (Yens) resembling Martian surfaces, and Mount Hood, Oregon, with icy regolith mixtures similar to lunar permanently shadowed regions.
• These sites allowed for testing the leg's ability to detect complex textures, layering, and varied ice content in the regolith.
• The NASA-funded project "Lassie" involves a team of planetary scientists, roboticists, and human-robot interaction researchers to facilitate planetary science discoveries.

"One is at Wense in New Mexico which is a dune field that possess very similar features such as the sand ripples and the crusty surfaces that has been observed by the Opportunity rover on the Mars."

Detecting Icy Regolith and Bioactivity Through Force Signatures [16:16]

• At Mount Hood, force drops and ductile failures were observed when the leg penetrated icy regolith, differing significantly from homogeneous sand. These signatures can infer ice content and binding forms, crucial for understanding lunar water origins.
• At Yens, distinct force signatures were detected between arid, actively migrating dunes and stabilized parabolic dunes with vegetation.
• A hypothesis suggests that biopolymers (EPS) from bacteria activity in more biodiverse areas cause a more ductile, leathery force signature in the surface crust, potentially revealing signs of past or present biological activity on other planets.

"So the implication of that is by looking at these force responses and the resilience of the crust signature, we can potentially reveal signs of either past or present bio activity on different planet."

Enabling Legged Robots to Walk and Sense Simultaneously [21:16]

• To gather data through every step, the sensing capability needs to be integrated into legged robots that can walk dynamically while maintaining high-precision force sensing.
• The "Spirit" robot, a quadruped with low gear ratio joints, was used to develop a sensing gait that maintains measurement accuracy.
• A static gait, where the robot supports itself with three legs and moves one at a time, was initially employed to minimize inertial effects and validate force measurements against controlled sand and crust surfaces.

"So the first thing that we did is we started to develop a sensing gate for the robot to walk while still maintaining the sensing accuracy."

Real-time Regolith Mapping and Traversal Risk Estimation [23:26]

• Initial results showed the robot could distinguish sand regions from other materials by measuring force per unit depth with each step.
• Deploying the robot in field sites like Yens and Mount Hood allowed for real-time regolith property mapping during locomotion.
• By analyzing force-depth curves, the robot detected variations in sand compaction and surface crusts, creating high-resolution maps of regolith strength.

"So you can see that actually the robot can quite effectively distinguish the sand region where it steps into the soft material."

Predicting Rover Traversal Risk from Regolith Properties [27:17]

• The second challenge is relating sensed regolith properties to the traversal risk for wheeled rovers, which is not always linear.
• Understanding the failure mechanisms of robots on sand, which can behave as either a solid or a fluid, is key to predicting failure points.
• Experiments in fluidized beds with various animals and robots revealed that locomotion success is tightly related to the "yield stress" behavior of the sand, where a threshold force determines whether the sand solidifies or behaves like a fluid.

"So the key challenge here is we actually don't have a equations at the level of Navis Stokes equation to relate these microscopic particle dynamics with B responses that is going to quickly impact our robot mobility."

Locomotion Model Based on Yield Stress and Sand Solidification Depth [30:52]

• The robot's speed during locomotion is tightly related to the yield stress behavior of the sand.
• When the applied force is below the yield stress, the sand solidifies; above it, the sand yields and flows like a fluid.
• The depth at which sand solidifies to support locomotion can be used to predict robot mobility, as demonstrated by experiments with various animals and robots collapsing onto a single curve when plotting speed against sensed solidification depth.

"So one key thing that we figure out is that the locomotion success is tightly related to the yield stress behavior of the sand."

Field Validation of Traversal Risk Estimation and Collaborative Exploration [33:06]

• A locomotion model based on sensed yield strength was validated in lab experiments, predicting robot speed based on its physical specifications.
• This model was then applied to larger rovers in field analog missions at NASA's lunar regolith test bed and the Yens field site.
• The "Lassie" project aims to create a collaboration between legged robots (scouts) and rovers for safe traversal, real-time regolith mapping, risk estimation, and failure recovery.

"So with this lab validation we then predict instead of this prototype robot we make this prediction on larger rovers and we validate our model in our two field analog machines."

Real-time Regolith Mapping and Risk-Aware Path Planning [35:43]

• In the NASA lunar regolith test bed, a scout robot created a real-time terrain strength map by sensing the ground during its zigzagging path.
• This map was used to estimate traversal risks for different rovers based on their payload capacity, guiding path planning to avoid high-risk areas.
• At the Yens field site, a naive path for a rover resulted in it getting stuck, while a risk-aware path, informed by the scout robot's map, allowed the rover to successfully navigate and reach its goal.

"So the robot walks around and makes the regular sensing through every step and then we can plot the training strength map from the data."

Collaborative Failure Recovery with Scout Robots [42:45]

• The research also explores using scout robots with telescoping arms to assist rovers that get stuck.
• Initial tests showed scout robots could dock with a struggling rover and use their arms to push or pull, aiding in failure recovery.
• This collaborative approach aims to enhance science return and prevent catastrophic failures during planetary exploration.

"And this is actually the second part of our luster project ma mostly done by Cindy and Mark's group where we wonder if the rover does get stuck if unexpected failure does occur can we use the scout robot to land arm to help them get out of trouble."

Calibration and Environmental Influences on Sensing [44:41]

• Calibration of proprioceptive sensing involves joint-level calibration for torque estimation and appendage-level calibration for sensor accuracy in various configurations.
• Dynamic locomotion introduces inertial effects that can impact force sensing accuracy, particularly for detecting surface textures and layering.
• Environmental conditions like wind, humidity, and temperature can influence results, highlighting the need for high temporal and spatial density data to understand these dynamic changes.

"What we have discovered, we just submitted our paper to Ukraine last week or the two weeks ago is that we found uh when we increase the gay speed uh the strength of the sand can still be picked up relatively reasonably."

Addressing Failure Modes and Optimizing Design [56:00]

• Failure modes such as slippage, stick-slip behavior on icy surfaces, and extreme sinkage in soft sand are investigated.
• The research explores how foot design, touchdown angle, speed, and trajectory influence interaction with the environment and affect mobility.
• Optimizing the balance between sandproofing, waterproofing, and heat dissipation for robots operating in extreme thermal environments is a critical design challenge.

"So essentially, you are stepping in, make ice sheet, break it, fresh snow again, make another ice sheet, break it. It was terrible, right?"

Future Directions: Sensing, Manipulation, and Decision Making [59:32]

• Future work involves investigating how different geometries and trajectories of robotic appendages can optimize sensing informative for terrain property estimation.
• The goal is to adapt the sensing methodology to various platform-specific risks, including wheeled versus legged locomotion.
• Research is also focused on developing intelligent decision-making algorithms for robots to balance reward and risk, optimize path planning, and extend mission lifetime, moving beyond human-controlled decision-making on Earth.

"So that's something that we are currently doing as of this moment. So thank you for the you predicted my uh you predicted our move essentially."