Today we're introducing our next generation humanoid battery for F.03. Like our actuators, vertically integrating our battery system is critical to Figure’s success. Engineered in-house and manufactured at BotQ

We're scaling up BotQ, our high rate manufacturing facility:. → Software Test Eng.→ Electrical Test Eng.→ End of Line Lead.→ Equipment Eng.→ Technician.→ MFG Eng. Consider joining us:.

Today, we're sharing a detailed write-up of the F.03 battery development. We discuss battery architecture, thermal runaway, and manufacturing.

Anti-Propagation & Flame Containment System. The pack includes safety measures to stop thermal runaway from spreading to nearby cells and to contain flames. This is achieved using a thermally insulative potting compound and rapid heat distribution

Active Cooling System. We've created an active cooling system with integrated cooling components in the die casting. This reduces thermal resistance and local heat, allowing fast charging with basic forced convection cooling

Structural Battery. Our structural battery pack features a high-strength enclosure made of:. → stamped steel.→ die-cast aluminum.→ structural adhesives. This enables it to withstand a 1-meter drop onto concrete from any angle

Advancements across 3 generations of battery technology has resulted in:. → 78% cheaper.→ Higher safety and reliability.→ 2kW fast charge with active cooling.→ Significant energy density improvements

If you're interested in shipping humanoid robots to the world, at scale, consider joining the Figure team. We are hiring today across Helix, BotQ (Manufacturing), Mechanical & Electrical Engineering, Software and Fleet Operations.

Today we posted a long-form write-up of Helix solving Logistics. We've made meaningful advances in model architecture which we walk through at length here:.

You'll find a few mistakes but they're declining as we feed more data and training time into the models. What is great about the models is they can re-plan after a mistake (e.g. robot misses grabbing a bag and will retry until it's successful).

In addition to model improvements, we also saw strong gains from scaling training data. As we scaled from 10 to 60 hours of training data we saw the time/package drop from 6.3 to 4.3 seconds and barcode scanning accuracy improve from 88% to ~95%

Many of the gains shown here were enabled by improvements to Helix’s System 1 policy. Helix now incorporates upgrades in temporal memory and force feedback. For example: the robot can now feel when it first touches an object

In the last few months, we've advanced our Helix neural network, which in Logistics is now capable of:. → New package types.→ 20% faster package handling.→ 35% higher barcode scanning success.→ Adaptive behaviors such as patting down packages.

It would've been impossible to do this with traditional robotic approaches (e.g. coding). This is a perfect problem for a neural network like Helix - notice the variety of objects and the subtle behaviors

Watch Helix's neural network do 60 minutes of uninterrupted logistics work. Helix now incorporates touch and short-term memory and it's performance continuously improves over time

RT @adcock_brett: Figure is officially 3 years old 🎉. Over the last 3 years, we’ve designed three generations of robots, deployed commercia….

RT @adcock_brett: On Friday, Figure completed a 20-hour run of back-to-back shifts on the BMW X3 production line!. We’ve been running 10-ho….

RT @adcock_brett: The Figure team wrapped up a productive two-week visit at BMW Group Plant Spartanburg last month. We optimized the proces….

RT @adcock_brett: At 9am ET this morning, my litigation counsel took action to protect and defend the reputation and the future of Figure f….

Consider joining our Helix AI team to help scale intelligence to millions of robots:. > Training Infra.> Manipulation Eng.> Large Scale Training.> Large Scale Model Evals.> Reinforcement Learning. Apply here: