Why are miso-scale robots important?

Miso-scale robots are crucial for applications requiring navigation in confined spaces, such as collapsed buildings or tight tunnels, and for performing precise operations like in precision agriculture. They offer a balance between mobility and the ability to handle larger obstacles.

What is the main challenge for miso-scale robots?

The primary challenge is operating in a 'noise-dominated regime.' At this scale, robots interact with multiple objects of comparable weight, making terrain reaction forces unpredictable and difficult to model.

Can robots achieve reliable locomotion without feedback ('computational intelligence')?

Yes, drawing an analogy from information theory, robots can achieve reliable locomotion through 'morphological intelligence.' This involves exploiting physical laws and morphological constraints, such as leg redundancy, to passively handle environmental perturbations.

How does having more legs improve locomotion?

Increasing the number of legs, a form of morphological redundancy, leads to more robust locomotion over complex terrains by consistently generating thrust and reducing the variability in arrival times. This principle is demonstrated with robots ranging from six to sixteen legs.

Can morphology influence speed as well as robustness?

Intuitively, longer legs are associated with speed. However, the research shows that controlling multi-legged robots differently, inspired by centipedes' terrestrial swimming, can achieve significantly higher speeds on flat ground compared to leg-driven locomotion.

How can robots go beyond biological inspiration?

By applying physics principles, robots can explore locomotion strategies not found in nature. For instance, asymmetrical designs or even limbless robots curling into a helix can achieve faster or more energy-efficient movement than biological counterparts.

What is the relationship between morphological and computational intelligence?

There may be a functional equivalence, where the combined effect of morphological intelligence (physical design) and computational intelligence (feedback control) determines performance. Different distributions of these two can lead to emergent, on-demand embodied intelligence tailored to specific constraints like speed and robustness.

Key Moments

Stanford Robotics Seminar ENGR319 | Spring 2026 | Mechanical Intelligence in Locomotion

Stanford Online

Education5 min read52 min video

Apr 22, 2026|162 views|18

Stanford Stanford Online Robotics

Save to Pod

Key Moments

TL;DR

Beyond 100 legs, adding more does not linearly increase a robot's speed, but asymmetry in both form and function unlocks surprising gains.

Key Insights

A significant research gap exists for "miso scale" robots, weighing approximately 1 kilogram, between the common micro (<1g) and macro (>10kg) robots.

Morphological intelligence, analogous to digital signal redundancy in information theory, allows for reliable locomotion over noisy terrains without feedback.

A 16-legged robot without feedback achieved similar performance to an 8-legged robot with feedback at 0.5 Hz, suggesting functional equivalence between morphological and computational intelligence.

Centipedes navigate terrestrial environments more like viscous swimmers than inertia-driven ones, with their locomotion dominated by consistent thrust generation against drag.

As limb size decreases, sensing modality should shift from legs to the body; for instance, torque distribution on the body predicts granular media depth for skink robots, whereas for quadrupeds, leg torque distribution is key.

Asymmetry, rather than symmetry, in both morphology and computation can lead to significantly faster locomotion, with one asymmetrical quadruped design achieving 50% faster movement than conventional symmetric designs.

The overlooked niche of the miso-scale robot

The seminar introduces "miso scale" robots, defined as those weighing around 1 kilogram, filling a critical gap between micro (<1 gram) and macro (>10 kilogram) robots. This scale is crucial for applications requiring navigation in confined spaces, such as collapsed buildings or for threat detection systems (projected $21.5 billion market by 2028), and precision agriculture (projected $21.9 billion market by 2031) where large machinery would cause damage. Unlike micro-robots that interact with one object at a time and macro-robots that can treat many objects as a continuous medium, miso-scale robots must interact with approximately ten objects of comparable weight, leading to a "noise-dominated regime" where terrain interaction forces are unpredictable. This challenge stems from the difficulty in quantifying terrain dynamics and uncertainty, limiting the development of effective miso-scale robots capable of reliable locomotion.

Morphological intelligence as a solution for noisy terrains

Drawing an analogy from information theory, the speaker proposes "morphological intelligence" as a means to achieve reliable locomotion in noisy environments, particularly for multi-legged robots. Just as redundancy in signal coding allows for reliable digital transmission over noisy channels without feedback, morphological redundancy – having many legs – can enable predictable locomotion. This principle suggests that if a robot possesses sufficient morphological redundancy, it can navigate complex terrains without relying on active computational feedback. This concept challenges the conventional assumption that computational intelligence is a prerequisite for step-driven locomotion. Experiments with robots ranging from six to 16 legs demonstrated that while speed on flat ground is similar, the 16-legged robot maintained consistent velocity and reduced the variation in arrival times on complex terrains, offering a guaranteed locomotion capability.

Centipedes leverage 'terrestrial swimming' for speed

While multi-legged robots excel in robustness, they are often slower than conventional robots. To address this, the research explores how morphology can be adapted for speed, inspired by centipedes. Centipedes, despite their numerous legs, can achieve significantly higher speeds on flat ground than when navigating complex terrains. Their locomotion on flat ground is characterized as "terrestrial swimming," closer to viscous-driven movement than inertia-driven movement, similar to eels. This is quantified by a "coasting number" significantly less than one, indicating that thrust generation against environmental drag is more important than inertial forces. The coordinated body and leg movements of centipedes are key to this thrust generation, allowing them to achieve speeds up to three times faster than leg-driven locomotion, aligning with a many-to-many mapping between morphology and performance.

Balancing speed and robustness: functional equivalence of intelligence

The research investigates how to achieve both high speed and robustness in robots. It was observed that a 16-legged robot with no feedback exhibited performance levels comparable to a 12-legged robot with feedback at 0.1 Hz, and further comparable to an 8-legged robot with feedback at 0.5 Hz. This suggests a "functional equivalence" between morphological intelligence (multiple legs) and computational intelligence (feedback control). This observation leads to the concept of "emergent embodied intelligence," which results from different distributions of morphological and computational intelligence. The goal is to co-design these two forms of intelligence on demand, allowing robots to achieve desired levels of speed and robustness by strategically distributing intelligence between their physical form and their control system.

Adapting sensing to morphology: the skink robot example

To meet the challenge of controlling unconventionally shaped robots and quantifying morphological benefits, the research introduces a "skink robot" with tiny legs and an elongated body, aiming to combine the advantages of snake and quadruped robots. The control strategy adapts based on terrain complexity, transitioning from standing waves to traveling waves as the robot moves into deeper granular media. Sensing for terrain estimation also adapts: for the skink robot, torque distribution on the body predicts granular media depth with 90% accuracy, whereas for conventional quadrupeds, leg torque distribution is indicative. This highlights a transition in sensing modality from legs to body as limb size decreases, enabling adaptive locomotion.

Transcending biology for locomotion

While biology provides inspiration, there are benefits to transcending its limitations. For instance, increasing the number of legs in multi-legged robots yields diminishing returns in speed, saturating around seven pairs of legs for top speed. However, by breaking away from biological constraints, such as moving sideways instead of always forwards, speed can increase linearly with the number of legs. In extreme cases, limbless robots can curl into a helix to roll like a wheel, achieving extremely efficient and fast locomotion. This underscores the importance of understanding when to move beyond direct biological mimicry, but also highlights the challenge of exploring the vast dimensionalities of multi-legged robots without biological reference.

Asymmetry unlocks superior locomotion

Physics-based exploration of locomotion, particularly in hexapods, reveals that asymmetry can be more beneficial than symmetry for achieving high speeds. While biological systems often rely on symmetrical movements for straight-line locomotion, a graph optimization approach showed that asymmetric patterns – turning clockwise for three-quarters of a cycle and counterclockwise for one-quarter – lead to faster movement. In fact, this optimization allowed for two motors to be fixed, enabling the design of asymmetrical quadrupeds with three legs on one side and one on the other, achieving 50% faster locomotion than conventional symmetric designs. This asymmetry in both morphology and computation is key to reliable locomotion, particularly in challenging conditions like snow.

Mentioned in This Episode

●Concepts

●People Referenced

Miso-Scale Robot Locomotion Principles

Practical takeaways from this episode

Do This

Embrace morphological intelligence through sufficient redundancy (e.g., more legs) for robust locomotion on complex terrains.

Explore body-driven locomotion inspired by centipedes for increased speed on flat surfaces.

Adapt sensing modalities to morphology; for tiny legs, sense via the body; for long legs, sense via the legs.

Utilize physics and graph optimization to explore full contact state sequences, even those that are asymmetrical.

Consider asymmetry in both morphological and computational intelligence for optimized locomotion.

Avoid This

Do not assume a one-to-one mapping between morphology and performance; explore many-to-many relationships.

Do not be constrained by biological rules; consider transcending them for novel locomotion principles (e.g., sideways movement, rolling).

Do not strictly adhere to contralateral symmetry in robot design if asymmetry yields better performance.

Do not ignore the potential for morphological intelligence to provide speed, not just robustness, by adapting control strategies.