Last week, speculation raged about Alphabet Inc.’s potential sale of Boston Dynamics Inc., whose humanoid Atlas robot and quadruped LS3 robot have been the subject of viral videos on YouTube.

Why would Google Inc.’s parent dump such a well-known robotics company? Is this a setback for Google or Boston Dynamics, or is the move good business sense?

What are the implications for other big robotics projects? How does one balance innovation with returns? Let’s take a look at how Google and Boston Dynamics’ team-up unraveled.

1. A promising partnership stumbled
As with many mergers and acquisitions, a union that might look good on paper is often more difficult to manage than expected in terms of corporate cultures and goals.

Waltham, Mass.-based Boston Dynamics was spun out of MIT in 1992, and Google acquired Boston Dynamics in 2013.

Google’s Replicant robotic initiative was plagued by leadership turnover and had difficulty getting teams in Tokyo, San Francisco, and Boston to work together, according to Bloomberg.

Aaron Edsinger, director of robotics at Google in San Francisco, said he had tried to work with Boston Dynamics to create a low-cost quadruped but encountered “a bit of a brick wall,” according to documents obtained by Bloomberg.

Andy Rubin, former leader of Android, had hired 300 engineers but then left Google in 2014, and there has been leadership turnover since then.

2. The Marines drop robot mule
In addition, the U.S. Marine Corps decided to pass on Boston Dynamics’ LS3 robotic “mule,” or “Big Dog,” after testing it.

The U.S. military, which is already using robots such as tracked PackBot (formerly from iRobot Inc.) for infantry support, said that the LS3’s gasoline motor was too loud. The contract was worth $10 million.

Boston Dynamics did eventually produce a quieter quadruped nicknamed “Spot,” but it lacked the payload capacity or autonomy to be helpful to soldiers and marines.

3. Humanoids are hard to get right
Several competitors used Atlas in last year’s DARPA Robotics Challenge, which highlighted the shortcomings of then-current humanoid robots. They were slow, unstable, and imprecise, a long way from the desired search-and-rescue functionality.

HUBO, the winning robot from the Korea Advanced Institute of Science and Technology, was not based on Atlas.

Despite the current lack of agility and autonomy, R&D into humanoid robots is likely to continue because of their perceived potential to cooperate in areas designed for humans.

For instance, NASA last fall awarded two Valkyrie robots to research teams at MIT and Northeastern University. Each team won $250,000 to continue developing the humanoid R5s, which could someday be used in space exploration.

In February, the latest version of Atlas (which includes 3D-printed components) got a lot of attention with a video demonstrating its improved walking ability and stability despite being poked with hockey stick.

The popularity of Boston Dynamics’ robots in the media may also have hurt the partnership, since Google didn’t want to be associated with any potential backlash against “killer robots” or job-displacement fears.

“There’s excitement from the tech press, but we’re also starting to see some negative threads about it being terrifying, ready to take humans’ jobs,” wrote Courtney Hohne, a Google communications director, in an internal e-mail.

4. Google is focusing on other things
Created last year, Alphabet has been focused on developing products with more immediate revenue prospects. It rolled Replicant into the $500 million Google X and reassigned researchers.

“We as a startup cannot spend 30 percent of our resources on things that take 10 years,” said Jonathan Rosenberg, senior vice president of products at Google and an advisor to Alphabet CEO Larry Page.

Not only is Alphabet working on self-driving cars, but it’s also investing in artificial intelligence. Google has filed for a patent to program “personalities,” and its AlphaGo program beat a human master at the Go board game four out of five times.

Like IBM Watson and others, Google hopes that its AI will prove helpful to medical research, machine vision, autonomous navigation, and human-machine interactions. There’s a lot of work yet to be done, and it’s not yet clear whether Alphabet will invest in another robotics hardware company.

5. Don’t worry; Boston Dynamics should do just fine
Boston Dynamics still has contracts with partners including the Defense Advanced Research Projects Agency (DARPA), Sony Corp., and others, and its robots could find new users.

Several sources have named Amazon.com Inc. and the Toyota Research Institute, among others, as possible buyers.

Amazon’s Kiva Systems division produces tens of thousands of robots for its warehouses, and Toyota Motor Corp.’s bicoastal institute is working on self-driving cars and humanoid “partner” robots.

Spot may antagonize a real dog and be too small to carry commercial loads, and Atlas has a long way to go to gain intelligence and endurance, but both could eventually find their way into commercial sites and homes.

However, Boston Dynamics would have to compete with other companies already working on affordable robots for those uses. The company may have to look elsewhere for partners that support its longer-term research.