With its advanced design, capabilities, and potential, Boston Dynamics’ Atlas sets a new standard for humanoid robots. Atlas is very interesting because it moves, controls, and makes decisions in a strangely human way. It can negotiate rugged terrain and obstacles, has a stereo vision with range sensing, and can manipulate objects on its path. The robot has 3D-printed parts to save space and weight, giving it a high strength-to-weight ratio. Atlas is a research robot platform and is not available for sale for commercial purposes, unlike Spot, the robot dog (Read Here).
Let’s dive into the details of the robot and understand its history, purpose, capabilities, and issues.
History of Atlas
Petman, a robot built by Boston Dynamics, was the base for Atlas. Petman would walk, squat, and do pushups, and could change their body temperature, humidity, and sweat on demand.
On July 11, 2013, the first version of Atlas was shown to the public. It was made for the Defense Advanced Research Projects Agency (DARPA) to be a robot that could help in a disaster. DARPA is responsible for the development of emerging technologies for use by the military in the US. Atlas was created for the DARPA Robotics Challenge (DRC), held between 2012 and 2015. The goal of the DRC was to speed up the development of advanced hardware, software, sensors, and control interfaces for robots so that they can help humans deal with natural and man-made disasters in the future. Seven groups of scientists from top schools like MIT and Virginia Tech were each given a model to work on. They had to program their software to overcome the obstacles of the tournament.
Characteristics of Atlas version 1
When first introduced to the public, Atlas weighed around 330 lb (150 kg) and was 6′ 2″ in height. The body was made of aluminum, steel, and titanium, which made it strong and impact resistant. The onboard computers could collect data, monitor sensors, and communicate with a remote user. It had a stereo vision and could do complex movements.
In the finals of the DRC 2015, the robot was able to complete all the given 8 tasks, namely
- Drive a utility vehicle at the site.
- Travel dismounted across the rubble.
- Remove debris blocking an entryway.
- Open a door and enter a building.
- Climb an industrial ladder and traverse an industrial walkway.
- Use a tool to break through a concrete panel.
- Locate and close a valve near a leaking pipe.
- Connect a fire hose to a standpipe and turn on a valve.
Atlas is a research platform robot, and its only purpose is to push the limits of whole-body mobility. The Atlas that we know today is a complete evolution from what was first unveiled. It is lighter, faster, and can do a lot of complex movements like parkour, backflips,
|Weight||196 lb (89 kg)|
|Height||5ft (1.5 m)|
|Degrees of Freedom||28 hydraulic joints|
|Navigation Sensors||LiDAR sensors, |
Inertial Measurement Unit (IMU)
It features one of the world’s most compact hydraulic systems for movement. A custom battery, valve, and power unit deliver high power to any of its 28 hydraulic joints, enabling it to do impressive mobility feats. The advanced control system enables highly diverse movements for the robot, and the complex algorithms allow it to do complex interactions with its environment with its whole body. Apart from that, the 3D-printed parts give it a high strength-to-weight ratio for summersaults and leaps.
Capabilities of Boston Dynamics’ Atlas
Atlas has three important components based on which it controls its motion and makes decisions, namely:
- Behavior Libraries: This enables the robot to be fed a template of a simulated similar environment on how it should behave rather than providing it with the exact scenario. This makes robotic movements more autonomous and lesser controlled by humans. Template motions are created by using trajectory optimization techniques and compiled into complex solutions.
- Real-time perception: Uses depth sensors and color cameras to generate point clouds of the environment and detect its surroundings and range of objects.
- Model predictive control: Uses models of the robot’s dynamics to predict how its motions will evolve over time and adjust accordingly. As an example, this controls how the robot should behave when carrying an object over various terrains and not tip over.
Recently, a video released by Boston Dynamics shows two Atlas robots on a parkour course. That was the first time we saw the robot doing a backflip. Although backflipping or parkour doesn’t hold any specific commercial use case for a robot, it gives a lot of knowledge on how a human body controls itself in any given environment. The new generation of Atlas runs on its perception rather than being controlled by a human directly. That means the robots’ movements aren’t pre-programmed for each new scenario. Instead, small templated behaviors are executed to get the desired results.
Scott Kuindersma, Team Lead of Atlas, said “Atlas’s moves are driven by perception now, and they weren’t back then. For example, the previous floor routine and dance videos were about capturing our ability to create a variety of dynamic moves and chain them together into a routine that we could run over and over again. In that case, the robot’s control system still has to make lots of critical adjustments on the fly to maintain balance and posture goals, but the robot was not sensing and reacting to its environment”
In another recent video, we saw the robot’s Real Time Perception and Model Predictive control abilities, which enabled it to locate and identify an object and then analyzed the best way to grab and carry it. It then arranged the planks to create a pathway to deliver the toolkit to its destination before finally ending with a backflip. The backflip consisted of a multi-axis spin just like a figure skater and was a complex one to achieve for the engineers working on the model.
Many organizations and institutions are working on various prototypes of Humanoid robots across the world. A few of them are available commercially, while others like Boston Dynamic’s Atlas are only research platforms. Let’s learn about a few humanoid robots around
- Ameca from Engineered Arts: Ameca is the world’s most advanced human-shaped robot representing the forefront of human-robotics technology. It is a development platform and can read emotions and interact. However, it cannot walk. It’s available for rent and purchase
- Astro (Apptronik): Astro is an upper-body humanoid robot. Designed to operate with humans, it has a state-of-the-art actuation, packed into a small form factor that can be put on any mobility platform.
- Beomni (Beyond Imagination): Beomni is remote-controlled by “human pilots” wearing virtual reality headsets and other wearable devices like gloves. It follows the movements of humans and also learns using its AI to self-perform the task in the future.
- OceanOne (Stanford Robotics Lab): An underwater diving humanoid robot, OceanOne, from the Stanford Robotics Lab is exploring shipwrecks. It can reach a depth of 1000 meters and assist humans in discovering different aspects.
- Waker (UberTech): With improved hand-eye coordination and autonomous navigation, Walker X, a humanoid service robot by UBTECH Robotics, is able to safely climb stairs, balance on one leg, and have precise hand coordination.
- Sophia (Hanson Robotics): Sophia can process visual, emotional, and conversational data to better interact with humans. Sophia has traveled across the world and once featured on the Cosmopolitan magazine cover page once.
Current Issues and future development
Although Atlas is a very versatile and powerful humanoid robot, its movements are still far from perfect. Being a research robot, the main aim of Boston Dynamics is to gradually perfect the movements of Atlas and achieve near-human-level intelligence. If the Atlas of the future responds to the same level of dexterity as humans, then the potential is nearly limitless. Atlas performed all the tasks in one go during the parkour shoot. Earlier, these movements were achieved in individual sets. According to Boston Dynamics, the parkour display gave them a lot of new learnings and their obvious way forward is to improve its software components to make it perform better.