Reflections on Amazon's big news
My thoughts on recent big news from Amazon: Amazon Robotics hiring key members from Covariant and UC Berkeley and the fully in-person policy.
Robotics - Reinforcement learning
A few weeks ago, Amazon announced a big news:
Amazon is hiring Pieter Abbeel, Peter Chen, and Rocky Duan and licensing Covariant’s robotic foundation models to advance the state-of-the-art in intelligent and safe robots.
Pieter Abbeel is one of the big names in the world of reinforcement learning (RL). In fact, I was lucky enough to take his intro to AI class during my PhD at UC Berkeley (the class was hands down the most polished and well-run course I've ever taken). Some fundimentals about RL can be found in his open source lecture channel on YouTube.
Similar to other learning-based methods, RL has been around for over a decade, leaving a promising impression on people's minds. The core idea is to let a robot learn a control policy to achieve a specific task by interacting with the world and trying repeatedly. However, learning an effective policy for a robot controller is no easy feat, and the success of RL has largely been limited to academic lab settings. None of the robots featured in the famous Boston Dynamics videos (before 2024) were controlled using RL-based controllers.
Despite the challenges, researchers have persevered in RL, and recently, more and more production-ready robots are using RL-based controllers rather than the traditional, model-based approach. One of the leaders in this field are the ANYmal robots, whose technology is closely tied to the research output from the RSL lab at RSL lab at ETH Zürich. (You can check out their impressive YouTube channel for related videos.)
Therefore, it will be interesting to see what RL, and learning-based controllers in general, can achieve going forward. Stay tuned!
Industry - Return to office: a double edged sword for the managers
Another big announcement from Amazon this week is that all employees are expected to return to the office starting from the beginning of next year.
I understand this will impact many people's current work-life balance and cause significant stress across the organization. Even though I personally go to the office almost every day already, my initial reaction to this news is still negative. After all, no one enjoys being "required" to do something. However, I later realized that this policy, while significantly impacting employees, may have even greater implications for managers. Essentially, this shift in remote work policies presents a double-edged sword for managers.
On one hand, managers now have the power to require employees to come into the office if they sense a need for closer supervision. This is quite different from the previous 3-day return-to-office policies, where managers had little influence or control over their employees during the two work-from-home days.
On the other hand, this authority to ask employees to come to the office also comes with greater expected responsibility for the manager. During the "flexible" remote work days, it was largely the employee's freedom and responsibility to maintain their performance. Now, this flexibility is still permitted, but has to be explicitly signed off by the manager. Therefore, the manager has become accountable for how each of their employees are balancing their work inside and outside the office. If the team's performance falls short of expectations, the manager has limited excuses unless all their employees are attending the office 5 days a week.
The simplest approach for managers would be to just require all employees to come to the office 5 days a week. However, based on the collective experience gained since the pandemic days, we know that a "5 days a week" model is not the most productive for most people. Therefore, managers who aim to maximize team performance will now have the lever to provide more supervision to some employees, while also sharing more the responsibility for their employees' performance.