Maximizing the value of peer review
This week, I want to share what I learned about Amazon's peer review process and a bit of inverse kinematics.
Industry - Writing feedbacks
At the end of the year, Amazon asks employees to send invitations to colleagues for feedback. This type of peer review process exists in most US companies, though the details may vary. Though this is a task that requires time and considerable thought, I find this process useful and want to share some lessons learned:
Writing feedback is offering help. Putting our thoughts about people “out there” might be intimidating at first. As someone who doesn't want to appear judgmental or arrogant, I initially find it challenging. However, as I write more feedbacks, I realize it's not about sharing my opinion of the person, but what I think can help them grow. This mindset helps me be more objective when giving feedback.
Feedback is all about growth. This might be a strong statement, but asking ourselves questions like “Is this something the person should spend time improving?” or “Will the person know what to do to improve after reading the feedback?” can help us be constructive. After all, we don't want to spend time writing things that won't bring meaningful results.
Asking for feedback is also about growth. Junior employees often ask, “Who should I ask for feedback? Will the feedback affect my performance evaluation?” Internally, there is no direct statement that the feedback is for performance evaluation. However, by using common sense, certainly there is going to be some level of influence if the manager is going to see the feedbacks and is also giving evaluation. Nonetheless, I remind myself that choosing feedback providers should be driven by “who can give me useful feedback?” rather than “who can give me positive feedback?” If everyone tries to give constructive feedback, it will sound more like “ideas” than “judgments,” resulting in mostly neutral comments.
A good feedback system needs protection. One of my favorite quotes from a Japanese anime says: 法が人を守るんじゃない, 人が法を守る (The law doesn't protect people, people protect the law). In the context of peer reviews, a good feedback system requires everyone's effort to do it “right.” With honesty, humility (to ask for feedback), critical thinking (about evidence to support feedback), and care for colleagues (to provide constructive feedback and help them grow), the feedback system can bring maximum value to employees. I'm fortunate to have benefited from the feedback I've received in the past two years, and I hope our system continues to provide value to Amazonians.
A few more interesting points about Amazon's review process:
Good feedback can be short. Each year, we send out around 10 feedback requests and receive roughly 10 invitations. For each feedback, we write two text boxes with a 60-word limit each. This forces us to be concise and makes it easy for the person reading the feedback.
“Superpower” vs “growth ideas”. Often, pointing out strengths is easier than weaknesses, even when trying to be “constructive.” However, we focus on giving growth ideas. Interestingly, the same quality can be both a person's superpower and a growth idea. Making something strong even stronger can sometimes bring more value to the person and the team than addressing less important weaknesses.
Robotics - Inverse kinematics [part 2]
This week I study a topic that is related to IK, i.e., null-space motion, shown in the GIF below. The robot can keep its end-of-arm-tool (EoAT) at the same pose while moving its joints.
To understand what is going on, we first need to define the relationship between joint-space (JS) motion and task-space (TS) motion. Similar to forward kinematics that maps joint-values to an EoAT pose, we have a equation that maps joint-velocity to EoAT velocity:
The matrix, J, is the geometric Jacobian of the EoAT with the specified joint-values. If the joints are moving but the EoAT is not, the joint-velocity has to be in the null space of J. Another way to look at this is by performing singular value decomposition for J, i.e.,
where columns in U (output singular vectors) that have non-zero singular values are feasible EoAT velocity directions. Each of them is achieved by a combination of joint motion described by its associated column in V (input singular vectors). On the other hand, columns in V that have zero singular value depict the combined joint motion that will not result in any EoAT velocity (so the EoAT will stay in place). These singular vectors are in the null space of J.