tron @uwaterloo | founding eng @PreloadAI
Toronto, Ontario
Joined June 2021
- Tweets119
- Following106
- Followers52
- Likes42.1K
Ryan Rahman ✈️ CoRL 2026 retweeted
Announcing the beta release of the Pi API. We’ve partnered with companies across diverse industries to adapt our models to their robots and applications, from harvesting strawberries to assembling servers. Now we’re opening access to more teams.
Predicting per-finger pressure signals from EMG alone. No glove or vision input at inference.
Building toward robots that learn how much force to apply, not just where to move. @PreloadAI
Ryan Rahman ✈️ CoRL 2026 retweeted
"The invention of the boat reduced the importance of swimming, but enabled the discovery of new lands."
Such an apt framing 🤌
Au contraire.
C'est une nouvelle ère qui s'ouvre pour les mathématiques.
Une ère où la démonstration formelle est largement automatisée et où l'accent sera reporté sur le développement de nouveaux concepts, nouvelles abstractions, nouvelles définitions, et nouvelles conjectures.
L'invention du bateau a réduit l'importance de la nage, mais a permis la découverte de nouvelles terres.
Barrier of entry is about to get a lot slimmer
@physical_int has big things planned 👀
Ryan Rahman ✈️ CoRL 2026 retweeted
We built an annotation layer for egocentric video.
Raw video in. Dense, time-aligned action labels out, plus a quality score with timestamped evidence.
Not just more data. Data you can actually trust for training.
@PreloadAI
Seeing isn’t sensing.
Building the nervous system for robots.
Force-control prototype in simulation.
@PreloadAI (YC F26)
Ego data was a great trend.
Now we’re about to show what’s possible with orders of magnitude less data.
We finally put together a video showing what we’ve been building at Preload (YC F26) 🧠🤖
We collect synchronized egocentric video + EMG + tactile data for robot manipulation, VLAs, and world models.
If you need robotics data, we can collect it end to end, at scale, starting as soon as tomorrow, and help integrate it into model training.
We’re also excited to work with companies exploring where dexterous manipulation can unlock new automation.
🎥 Launch video below
preload.ai
[email protected]
Ryan Rahman ✈️ CoRL 2026 retweeted
Got an open-arm bimanual setup and pour policy running in a day. Still misses the cup occasionally.
Next: same task from human ego video, no teleop.
@anvil_robotics
Built a Rocky-inspired robot in simulation, trained it to walk, then trained a diffusion policy to stack cubes using echolocation instead of vision.
No cameras, just range sensors.
Might have to try building it out IRL next 👀
Amaze amaze amaze.
Ryan Rahman ✈️ CoRL 2026 retweeted
Bimanual espresso making ☕️
Exploring task orchestration for long-horizon manipulation.
Ryan Rahman ✈️ CoRL 2026 retweeted
A tactile-augmented flow policy for bimanual dexterous origami.
The goal is autonomous paper-airplane folding on a 65-DoF — iros 2026
The system combines three ideas:
→ Mixture-of-Horizons action prediction
→ Force-aware MoE
→ Long-horizon subgoal planning
Ryan Rahman ✈️ CoRL 2026 retweeted
We made a robot fold laundry.
The interesting part wasn’t training a bigger model.
We focused on getting more out of the same policy through better data, post-training, inference, and execution.
+10% manipulation success
+17% inference throughput
1.3× effective control throughput
Here’s what changed:
Ryan Rahman ✈️ CoRL 2026 retweeted
Some recent thoughts.
ICL is useful. RL is useful. VLA, sim2real, world models, memory—all of these technologies can contribute to building general-purpose robots.
But we should remember that a general-purpose robot is still a robot: a physical product that ultimately needs to be deployed and deliver real value to customers.
Once you look at the actual challenges of deployment, it becomes clear that no single technique is enough. It requires a full system, with careful attention to detail across models, hardware, infrastructure, reliability, and operations. Even at the model level, it’s unlikely that any single idea will be the answer. VLA, RL, sim2real, world models, ICL, memory, and many other approaches will likely all play a role. The real challenge is combining them in the right way, so they complement each other and maximize the capability and utility of the overall system.
So IMO, there’s no need to overhype any single technical term. This industry would benefit from staying calm and patient, and focusing on building the holistic system that can actually work reliably in the real world and create real customer value.
Ryan Rahman ✈️ CoRL 2026 retweeted
the number of people who think VLA is chatgpt for robots is actually insane. you cannot just train on a bunch of internet videos and expect your robot to fold towels in a real warehouse.
every model deployed took hundreds of hours of teleop data, site-specific tuning, and engineers physically standing next to the robot fixing edge cases for weeks. the gap between a research demo and a production deployment is ungodly.
seriously, go buy an arm. wire it up. try to pick up a wet cardboard box in bad lighting. half this space has never touched a motor controller and it shows.
Ryan Rahman ✈️ CoRL 2026 retweeted
Robotics Research w/ certified bangers!
@DynaRobotics - Dyna-2
- Scaling laws for egocentric data closing cross embodiment
@GeneralistAI - Gen1.5
- In context learning emerged —> GPT3 Moment
@SkildAI - S1
- In-context learning, but at 10-minute horizon
What’s next week?