---
date: 2026-04-08
---

# 2026-04-08

### Morning — Reflections & planning

**Big picture: home robot by 2028.** Thinking about the goal of building a useful home robot by 2028. It's scary but necessary — this is the thing I need to do. Need to move away from action primitives ASAP, especially during the TRI project. The goal isn't just publishing — it's building something real.

**Slides workflow.** For presentations, probably use an AI slide builder like Gamma — feed it bullet notes and images from Claude.

**Peace & spiritual life.** Need to make time for peace. Try breathwork, check out the Calm app. Go to Khanega Sunday morning.

**Style.** Order more of the Levi polos and the button-up — they fit well. Add to default wardrobe rotation. Also need a better ironing setup.

### Today's plan

- Finish Panda calibration via Claude
- Assemble robot arm
- Work on 567 homework
- Diet: one Robowok meal, maybe a brownie
- Fitness: rows and volleyball
- Gym at 5, leave by 6
- **Goal: don't be rushed today. It's okay not to finish everything.**

### Sleep & energy

Need to break the late-night cycle and start sleeping early. For now, just recognize this is a temporary investment period into agents — but the pattern needs to change.

### Franka Panda — done with it

After weeks of fighting with the Panda, I'm done. It's the most frustrating part of my day. I hate using it. I'd rather be rebellious and just stop, or have my advisor get a RA to work on it for us. It's been too much — I'm walking away from it.

### PARA research idea — egocentric pre-training with point tracking

Idea: pre-train on large-scale egocentric video data using point tracking as the objective, then fine-tune for PARA. Could be a good direction to explore for the TRI internship. Worth investigating what egocentric datasets are available and how point-tracking pre-training would transfer to pixel-aligned action prediction.

### Claude agents / life infrastructure

Feeling a bit overwhelmed by how powerful the Claude agents setup is. There's a pull to immediately streamline everything in life with it. It's a good direction, but important to be patient — don't try to automate and optimize everything at once. Let the system grow organically. The KB and agents are an investment that compounds, not something to force overnight.

### Evening — Recovery & recollection

**What happened:** Stayed up too late, scrolled too much, ate too late, didn't respect the balance between productivity and rest. Led to a bad cascade of feeling bad. That's okay — recognized it, did some meditation, feel more like myself again. Going to volleyball now, which will be fun.

**Lesson:** Better respect the balance going forward. The burnout cascade is predictable: late night → scrolling → late eating → feeling bad → more of the same. Catch it earlier next time.

### Panda update — softening

Talked more about the Panda — actually, working with it is extremely exciting when it works. Might ask advisor (US) to get a master's student to help with the Panda experiments. Also not convinced it's strictly necessary since we already have real robot experiments on a smaller arm. Open question, not a hard no anymore.

### Robot arm — faster progress than expected

The custom arm build is going faster than expected. Printed the first three links and they have similar proportions to the ARX arm (our target). Compelling progress. Realistic to have another robot arm built within a week, especially since we bought a second 3D printer for parallel printing. Open design questions: might want to double-joint or add a second motor to the base.

### USC vs Bay Area vibe

Interesting being at USC vs somewhere in the Bay. USC is so much more aesthetics and vibes focused. Different energy.