Decoding SPOT: How Legged Robots Work

If you’re into robotics, you’ve probably heard of SPOT, the yellow, quadruped robot developed by Boston Dynamics. While for many, it might seem like a recent entrant into the robotics space, its roots stretch back over 30 years when Marc Raibert and his team started researching legged robots at the Leg Laboratory at MIT and CMU. In 1992, Marc founded Boston Dynamics, which was essentially an evolution of the research he led. The initial legged robot projects were funded by DARPA, the defense research agency of the US, which led to the development of massive, hydraulically powered robot dogs — namely BigDog in 2005 — and over the decades, this project evolved into a compact, battery-powered, electrically actuated platform: that’s SPOT.

Today, SPOT weighs around 32.5 kg (71.6 lbs), can carry a 14 kg (30 lbs) payload, runs for about 90 minutes on a full charge, and achieves speeds up to 1.6 m/s (3.5 mph). It’s not the only quadruped robot on the market—companies like Unitree Robotic (with their Go1 series) and researchers like MangDang (with their Mini Pupper) have developed similar platforms. SPOT, however, remains the gold standard when it comes to field-deployed, rugged-legged robots for industrial and commercial use.

So, why are these legged robots seeing growing adoption, even as wheeled mobile robots dominate warehouse floors? Legged robots thrive in places that are harsh, uneven, or inaccessible for humans or wheeled robots: deserts, forests, chemical plants, nuclear power stations, offshore oil rigs, and underground mines. In such high-risk zones, SPOT’s key value proposition boils down to: terrain adaptability and inspection.

In fact, most of SPOT’s commercial deployments today revolve around industrial inspection projects. For example, BP uses SPOT on oil rigs to monitor equipment and check for gas leaks — especially methane leaks, which are both dangerous and common on oil rigs. NASA’s Jet Propulsion Laboratory has tested SPOT in underground caves here on Earth, while the LA City government and New York Fire Department have adopted it to support public safety and emergency response. With each deployment, organizations gain the ability to collect frequent, high-fidelity data from high-risk environments and scenarios for human inspectors. Given the rapid pace of adoption of these legged robots if you are working in the robotics space or are planning to build your career in this domain,  you'll need to understand how legged robots function — both in terms of system architecture and field operations.

Lets first uncover the system architecture of this robot. Here we will learn the core systems that power modern legged robots.

System Architecture

SPOT’s system architecture consists of four key subsystems: Perception, Locomotion, Navigation, and Motion Control.

Perception

Perception is how SPOT senses the world around it. Onboard, it carries a 360-degree stereo camera array, depth cameras, an IMU (Inertial Measurement Unit), and an optional LiDAR unit. Using sensor fusion algorithms like Kalman Filter, SPOT combines the data from these sensors and builds a detailed 3D map of the environment. It is this perception layer that allows it to detect obstacles, estimate terrain roughness, and track SPOT’s own pose and orientation. Without this layer, the robot would effectively be “blind.”

Locomotion

Once SPOT builds a detailed understanding of its environment, its next critical task is retaining balance. This challenge is unique to legged robots because, unlike wheeled robots — which have a naturally stable base — legged robots must continuously retain balance as part of their core operation. That's where the locomotion system comes handy.

Locomotion refers to how SPOT interacts with the ground while maintaining balance. Each of SPOT’s four legs is equipped with force sensors at the foot tips and 12 high-torque brushless motors: three per leg: hip abduction/adduction, hip flexion/extension, and knee flexion/extension. These sensors measure how much force each foot exerts on the ground, allowing the system to detect slip, adjust posture, and recover from disturbances. 

The holy grail of legged robots lies in their ability to sense the terrain profile, understand the forces at play, and decide on one of several possible movement strategies, also known as gaits, at every moment. What do I mean by that? 

Well, SPOT has four legs, and at every step, it must determine the most appropriate gait based on multiple factors, including terrain conditions. Each gait defines how many legs are on the ground and how many are in the air as the robot moves. Naturally, there’s a tradeoff. If the ground is stable, SPOT can walk with two legs on the ground and two in the air, allowing for higher speed. On the other hand, if it’s climbing a hill or crossing uneven terrain, it might need to keep three legs grounded for maximum stability, lifting only one leg at a time. You see? The goal of the locomotion system is to decide these movement strategies. And how does it do that? That’s where the next module — navigation and motion control — comes into the picture.

Navigation & Motion Control

Navigation is where SPOT figures out how to move from point A to point B while staying balanced and stable. Unlike wheeled robots that follow a planned route on flat ground, legged robots face a more complex challenge: they must constantly adjust how they move — not just where they go. That’s because each step involves deciding how many legs should stay on the ground and how many should lift at any moment, depending on the terrain profile. This leg movement pattern is called a gait — and SPOT has to select the right gait dynamically as it moves.

To do this, SPOT uses a combination of control techniques, including Model Predictive Control (MPC) and, more recently, Reinforcement Learning (RL) approaches.

Model Predictive Control (MPC) helps SPOT predict its next physical state and the terrain ahead based on current sensor data and control actions. For example, if at time t = 1 second the terrain is flat, the system might assume — based on sensor readings — that by time t = 3 seconds, the terrain will remain largely unchanged. This means the controller would likely select the same gait or leg movement pattern it used earlier when the ground was flat. But, here’s the catch: what if, at t = 3 seconds, the terrain has completely changed — say, the robot has stepped from grassland into a shallow river? You see, while MPC offers important benefits, it also has limitations. It tends to assume that the world behaves in a mostly linear and predictable way, which may hold inside controlled lab settings but often breaks down in the messy, unpredictable conditions of the real world.

That’s where Reinforcement Learning (RL) becomes valuable. RL allows SPOT to learn stable navigation patterns through trial and error. The robot is trained in a simulated environment, where it’s rewarded for maintaining balance, moving efficiently, and completing tasks, while being penalized for actions that lead to falls, slips, or wasted energy. Over time, the learned policy enables the robot to adapt effectively to complex, changing terrain profiles — often outperforming traditional MPC approaches, especially when faced with the uncertainties in real-world environments.

Once the navigation system has figured out the best path and movement plan, the motion control system takes over. This part of the system translates the high-level plan into precise motor commands, controlling the torque and speed across all 12 leg motors. The control loop runs at extremely high frequencies — hundreds of times per second — allowing SPOT to react in real time, adjusting to small slips, wobbles, or sudden changes underfoot, ensuring smooth, stable, and balanced motion.

Operation in the Field

Now, you might wonder: is SPOT fully autonomous?

Technically, it can be. SPOT integrates with ROS (Robot Operating System), allowing engineers to map an environment, place waypoints, and execute autonomous missions. But in reality, many industrial use cases, especially in high-risk facilities like oil rigs, power plants, or mines, require supervised or teleoperated modes for safety and accountability.

SPOT comes equipped with a custom-designed joystick controller paired with a tablet interface, giving operators real-time access to live camera feeds, sensor data, and system diagnostics. This setup allows users to manually steer the robot, set navigation goals, or seamlessly switch between autonomous and manual control modes, depending on the mission’s needs. 

Beyond this, Boston Dynamics provides a web-based interface and a robust software development kit (SDK), opening the door for advanced customization. With the SDK, developers can build mission-specific applications, integrate SPOT into broader operational workflows, or design specialized control modules for tasks like automated inspections, site mapping, or environmental data collection — extending the robot’s utility profile far beyond its out-of-the-box capabilities.

Energy Efficiency, Payloads & Customization

When it comes to real-world deployment, one of the most overlooked yet critical aspects of SPOT’s performance is energy efficiency. Unlike wheeled or tracked robots, legged robots expend significantly more energy per unit distance. This is because every step requires lifting and shifting the mass of the robot against gravity, coordinating dozens of micro-movements, and absorbing impact forces exerted by terrain — all of which cost energy.

SPOT runs on a 659 Wh swappable lithium-ion battery that offers roughly 90 minutes of continuous operation, depending on terrain, payload, and mission profile. For industries like oil and gas or mining, where downtime is costly, SPOT’s ability to hot-swap batteries in under a minute is a major advantage, keeping inspection missions running without lengthy recharge cycles. Battery management systems on SPOT constantly monitor temperature, load, and discharge rates to balance longevity and performance, ensuring the robot doesn’t overdraw power when navigating steep or rough terrain.

SPOT isn’t just an off-the-shelf product — it’s a customizable platform. Boston Dynamics offers a suite of official payloads:

• 360° camera systems for photogrammetry and site mapping.

• Thermal imaging sensors for detecting overheating equipment.

•  Gas detection sensors for identifying hazardous chemicals or methane leaks.

• Manipulator arms (Spot Arm) that allow SPOT to open doors, push buttons, or handle objects with precision.

  
  

Beyond Boston Dynamics’ own offerings, third-party developers have used SPOT’s extensive SDK to integrate specialized payloads: from radiation detectors in nuclear plants to machine-learning-driven visual inspection modules for construction sites. The robot’s 14 kg (30 lbs) maximum payload capacity, paired with its modular top-mounted payload interface, makes it adaptable across industries.

This combination of an energy-aware, modular, customizable platform is what transforms SPOT from a research prototype into a field-ready industrial robot. Whether it’s autonomously patrolling oil rigs, scanning construction sites, or exploring hazardous tunnels, SPOT’s energy systems and payload versatility are key to delivering meaningful, real-world value.

We cover SPOT and other industrial robots in much greater depth in our course Robotics Foundations. If you’re interested in diving deeper—into system design, control algorithms, or hands-on case studies—check out our comprehensive course on Robotics Foundations.

Cheers,

Prathamesh 

Disclaimer: This blog is for educational purposes only and does not constitute financial, business, or legal advice. The experiences shared are based on past events. All opinions expressed are those of the author and do not represent the views of any mentioned companies. Readers are solely responsible for conducting their own due diligence and should seek professional legal or financial advice tailored to their specific circumstances. The author and publisher make no representations or warranties regarding the accuracy of the content and expressly disclaim any liability for decisions made or actions taken based on this blog.