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

Here's a comprehensive summary of the YouTube video transcript, maintaining the original language and technical precision:

Key Concepts

• Terrain-Aware High Mobility Robots: Robots designed to understand and adapt to diverse and challenging planetary surfaces for exploration.
• Proprioception-Based Sensing: Utilizing a robot's own internal state (joint torques, motor currents) to infer external environmental properties, rather than relying solely on external sensors.
• Regolith Mechanics: The study of the physical properties and behavior of loose, unconsolidated materials found on celestial bodies like the Moon and Mars.
• Yield Stress: A material property indicating the minimum stress required to cause permanent deformation.
• Fluidized Bed: A laboratory apparatus used to simulate granular materials behaving like fluids by passing air through them.
• Gait: The pattern of limb movements a robot uses to walk or move.
• Scouting Robot: A smaller, agile robot used to explore and map terrain ahead of a larger, less mobile rover.
• Traversal Risk: The probability of a robot encountering difficulties or failing during movement across a specific terrain.
• Biopolymer EPS (Extracellular Polymeric Substances): Organic compounds produced by microorganisms that can bind soil particles and affect soil properties.

Terrain-Aware High Mobility Robots for Planetary Exploration

The video discusses the critical role of robots in planetary exploration, especially as missions to the Moon and Mars become more ambitious. The primary challenge is enabling robots to safely and reliably operate on diverse and complex planetary surfaces, which are often unconsolidated, rocky, and uneven. These conditions can lead to robots getting stuck, costing valuable scientific opportunities. The research focuses on developing robots that can better understand their physical environments and adapt their actions accordingly.

1. Sensing Environmental Properties Through Interaction

A key focus is on enabling robots to gather information about their environment, particularly the physical properties of regolith, through direct interaction.

• The Challenge of Visual Sensing: Visual sensors alone are insufficient to discern critical differences in regolith properties, such as compaction, which can vary in strength by over fivefold even with minor changes in packing fraction.
• Proprioception as a Solution: Humans and animals use their feet to gather rich tactile information while moving. The research proposes using proprioception-based sensing in robots, leveraging the high force transparency of direct-drive actuators. This allows robot legs to act as sensors, providing "free" and dense spatial information with every step.
• "Traveler" Leg Prototype: A single leg prototype named "Traveler" was developed using direct-drive actuators. This five-bar linkage leg can perform various sensing protocols, such as poking and scraping, and its force sensing capability allows for the inference of applied forces from motor current/torque.
• Lab and Field Validation:
  • Fluidized Bed Experiments: A fluidized bed was used to precisely control sand strength by varying airflow. Experiments showed that the leg could accurately measure the stiffness of the sand, with the slope of the force-depth curve decreasing linearly with reduced volume fraction.
  • Field Deployments: The "Traveler" leg was deployed in analog environments:
    • Wense, New Mexico (Dune Field): Mimics Martian sand ripples and crusty surfaces.
    • Mount Hood, Oregon (Icy Regolith): Simulates lunar regolith in permanently shadowed regions, featuring a mixture of ice and volcanic dust.
• Discoveries from Field Deployments:
  • Mount Hood (Icy Regolith): The leg detected distinct force signatures for different ice content and binding forms. A sudden force drop indicated surface ice rupture, followed by a ductile failure in cohesive soil with ice bridges. This allows for inferring ice content and binding mechanisms, crucial for understanding lunar water.
  • Wense (Biosignatures): In arid dunes with low bioactivity versus stabilized dunes with vegetation and increased bioactivity, the leg detected different crust failure signatures. Brittle breaks were observed in arid sites, while ductile, leathery, and elastic failures occurred in sites with higher bioactivity. This difference was linked to the presence of biopolymer EPS, which binds soil particles and creates cohesive bonds, suggesting that force responses can reveal signs of past or present bioactivity.

2. Locomotion Models for Predicting Traversal Risk

The research aims to develop locomotion models that predict the outcomes of robot-environment interactions based on sensed properties, enabling informed decision-making.

• Understanding Failure Mechanisms: A key question is why robots fail on sand. Sand exhibits both solid-like and fluid-like behavior, and the transition between these states is critical for locomotion performance.
• Yield Stress and Locomotion: The locomotion success is tightly related to the yield stress of the sand. When the applied force is below the yield stress, the sand behaves solidly; above it, it behaves like a fluid.
• Robot Mobility Model: A simple model was developed where robot speed is plotted against sensed solidification depth. This depth, which can be determined by matching the applied force to the material's yield force, exclusively dictates the robot's speed. This allows for predicting robot mobility and traversal risk based on sensed yield strength.
• Validation:
  • Lab Validation: The leg's sensing capability was used to poke into sand of varying stiffness. The locomotion model then predicted robot speed based on these properties, which was then experimentally validated against actual robot speeds in a fluidized bed. The model effectively captured the robot's failing trend and areas of difficulty.
  • Field Validation (LASSIE Project):
    • NASA Ames Lunar Regolith Testbed: A scout robot (Spirit quadruped) mapped the terrain strength in a simulated lunar regolith. The inferred strength map, generated from every step, closely matched the designed pattern. This map was then used to compute traversal risk for different rover payload capacities, guiding payload configuration.
    • Wense Field Site: The scout robot created a real-time regolith strength map. This map was converted into a traversal risk map. A comparison between a "naive" path and a "risk-aware" path showed that the risk-aware path, which avoided high-risk (soft sand) regions, successfully navigated to the goal without getting stuck, unlike the naive path.
• Collaborative Exploration and Failure Recovery: The LASSIE project also explored collaboration between scout robots and rovers. In one scenario, scout robots with telescoping arms were used to assist a rover that got stuck on a steep sand incline, demonstrating a capability for failure recovery.

3. Future Directions and Challenges

• Robot as a Scout: The vision is to use legged robots as scouts to identify scientifically valuable sites and map traversal risks for larger rovers, enhancing safe exploration.
• Calibration and Drift: Calibration is performed at the joint and appendage levels. While direct-drive motors offer good torque constants, dynamic locomotion introduces inertia and body motion that affect force sensing accuracy. Increasing gait speed can make detecting surface texture and layering less reliable. Over time, friction and wear can degrade performance, requiring recalibration.
• Optimization of Robot Teams: The optimal balance between hardening a large rover versus deploying multiple smaller scout robots for assistance and coverage is an ongoing area of research. The idea is to create teams of robots that can share information, collectively perform tasks, and provide redundancy.
• Environmental Influences: Environmental factors like wind, humidity, and temperature changes can influence surface properties. The research aims to correlate force changes with environmental conditions to understand dominant factors.
• Mechanical Design for Harsh Environments: Sealing robots against dust and water is crucial, but heat dissipation is also a significant challenge, especially in hot environments. Balancing these requirements is a key design consideration.
• Extended Lifetime and Efficient Planning: Robots need to make smarter decisions to balance reward and risk within battery and time constraints. This involves developing algorithms for autonomous decision-making based on incoming sensor data and changing environmental conditions, mirroring human decision-making processes.
• Diverse Failure Modes: The research investigates various failure modes beyond simple sinkage, including slippage, cohesive media (ice/snow) failures, and stick-slip phenomena. Understanding these modes is crucial for designing robust locomotion and interaction strategies.
• Geometry and Sensing: The geometry of the robot's appendage and its trajectory significantly impact sensing capabilities. Research is ongoing to understand how different geometries and movements can be optimized for sampling terrain properties and distinguishing between different material types.
• Manipulation and Real-time Optimization: The methodology for sensing regolith strength can be adapted for manipulation tasks and to assess traversal risk for different platforms (wheeled vs. legged) by breaking down contact surfaces into infinitesimal plates.

The overarching goal is to develop robots that can not only traverse challenging terrains but also actively sense, understand, and adapt to their environment, leading to more effective and scientifically rewarding planetary exploration.