Boston Dynamics Robots: From BigDog to Atlas

Three Decades of Robots That Refuse to Fall Down

Boston Dynamics was founded in 1992 as a spin-off from the Massachusetts Institute of Technology, where Marc Raibert had spent a decade studying dynamic locomotion — the ability of machines to walk, run, and balance using active control rather than static stability. The company's philosophy has been consistent: build robots that move through the physical world as capably as animals do. The results — BigDog, Atlas, Spot — have generated billions of views online and pushed the boundaries of what legged machines can achieve.

BigDog (2005): The Mule That Started It All

Funded by DARPA, BigDog was designed to carry 150 kg of gear over terrain too rough for wheeled vehicles. It weighed 110 kg, stood about 1 meter tall, and moved on four hydraulically actuated legs powered by a two-stroke gasoline engine. The engine was loud — a persistent criticism — but the locomotion was groundbreaking. BigDog could walk on ice, climb rubble-strewn slopes at 35-degree angles, and recover from hard lateral kicks without falling.

• Top speed: 6.4 km/h on flat ground
• Payload capacity: 150 kg
• Power: two-stroke single-cylinder engine driving hydraulic pump
• Sensors: LIDAR, stereo vision, joint position encoders, inertial measurement unit
• Notable: famous "kick test" video demonstrated dynamic balance recovery

LS3 and the Military Path

BigDog led to the Legged Squad Support System (LS3), a larger quadruped that carried 180 kg and followed soldiers using GPS and computer vision. The Marines tested it in field exercises in 2014. Ultimately, the military decided it was too noisy for combat operations. The project ended. But the technology it developed — particularly in terrain perception and dynamic gait control — flowed directly into later robots.

Atlas: A Humanoid That Does Parkour

Atlas, first unveiled in 2013 as a DARPA Robotics Challenge contestant, stands 1.5 meters tall and weighs about 89 kg in its latest hydraulic version. It uses 28 hydraulic joints and a battery-powered hydraulic system (replacing the earlier external power tether). Atlas has demonstrated backflips, parkour sequences, gymnastics routines, and manipulation of objects in unstructured environments.

In 2024, Boston Dynamics retired the hydraulic Atlas and introduced an all-electric version designed for real-world deployment in manufacturing facilities. The electric Atlas has a wider range of motion than the hydraulic version, with joints that can rotate 360 degrees. Hyundai, which acquired Boston Dynamics in 2021 for approximately $1.1 billion, plans to deploy the electric Atlas in its automotive factories.

Spot: The Commercial Quadruped

Spot, released commercially in 2020 at a price of $74,500, became Boston Dynamics' first revenue-generating robot. Weighing 32 kg, Spot can carry 14 kg of payload, navigate autonomously through mapped environments, climb stairs, and operate for 90 minutes on a single battery charge. Over 1,500 units have been deployed worldwide.

• Used by utilities for remote inspection of power substations
• Deployed at construction sites for progress monitoring and 3D scanning
• Oil and gas companies use Spot for routine inspections at offshore platforms
• NYPD's Digidog program tested Spot for hostage situations and hazmat scenarios
• NASA JPL tested Spot as a scout for cave and tunnel exploration

Spot's Sensor Suite and Autonomy

Stretch: Purpose-Built for Warehouses

Stretch, introduced in 2021, is Boston Dynamics' warehouse robot. It has a mobile base on wheels, a perception mast, and a large vacuum gripper on an articulated arm that can unload 800 boxes per hour from truck trailers. DHL deployed Stretch units in its North American distribution centers starting in 2023. Unlike Atlas and Spot, Stretch prioritizes productivity over versatility — it does one thing extremely well.

The Technology Stack Behind Boston Dynamics

The robots share common engineering foundations. Model Predictive Control (MPC) algorithms plan movements milliseconds into the future, adjusting thousands of times per second. Whole-body control integrates balance, locomotion, and manipulation into a single optimization problem. Computer vision systems built on deep learning identify terrain features, obstacles, and objects. Reinforcement learning, increasingly used alongside traditional control, has enabled more robust and adaptable behaviors.

• Hydraulic actuation provides high power density for Atlas and BigDog
• Electric actuation in Spot and new Atlas offers quieter, more efficient operation
• Custom battery management systems maximize runtime under dynamic loads
• Simultaneous Localization and Mapping (SLAM) enables autonomous navigation
• Real-time optimization runs at 200–1,000 Hz for balance control

Boston Dynamics has spent over three decades proving that legged robots can handle the physical world. The question is no longer whether these machines can walk, run, or jump. It is where they will show up next — and what they will do when they arrive.