Robotics
product
Humanoid robotics
Humanoid robotics moved from research demos to low-volume industrial production between late 2024 and May 2026 — Tesla and Boston Dynamics in factories, Figure on the BMW line, 1X taking consumer pre-orders, with VLA models (RT-2, π0, Helix) finally enabling task generalization beyond hand-coded behaviors.
What to watch next
Whether Helix-class full-body autonomy generalizes to unstructured homes (1X NEO, Figure 03), whether teleop data factories drop unit-economic costs below $50/hour, and whether Tesla Optimus V3 actually hits 1M units/year — the inflection from industrial pilot to mainstream tooling.
Key sub-ideas & techniques
- Vision-Language-Action (VLA) foundation models — End-to-end neural policies that combine VLM backbones with robot action tokens — RT-2, RT-X / Open-X-Embodiment, Physical Intelligence π0/π0.5, and Figure's Helix enable zero-shot generalization across embodiments and tasks. [source]
- Tactile sensing & dexterous manipulation — Multi-axis fingertip and palm sensors (3–5 mN sensitivity) fused with vision close the in-hand-manipulation gap — slip detection, contact-rich assembly, and delicate grasping become tractable. [source]
- Sim-to-real RL & whole-body control — Domain-randomized RL trained in Isaac Lab/Sim plus hybrid imitation pipelines now transfer zero-shot to hardware, with whole-body controllers handling balance and locomotion across embodiments. [source]
- Data factories & teleoperation scaling — Large-scale teleop pools (45–50 tutors per 100 robots) and at-home gig-economy recording have driven cost-per-tutor-hour down ~60% to ~$118 in 2026, accelerating data-hungry VLA training. [source]
- Form-factor convergence (26–32 DOF bipedal) — The industry has converged on ~26–32 DOF bipedal designs with 16–22 DOF tendon-driven hands, enabling vertical integration and shipment scaling toward 90K+ units in 2026. [source]
- MolmoAct2 (open VLA) — Open-weight Vision-Language-Action foundation model for bimanual humanoid manipulation, trained on 720 h of teleoperation, with public weights, dataset, and eval code. [source]
Current frontier
- Figure's Helix 02 (Jan 2026) shows unified full-body autonomy from a single VLA — dishwasher loading, bottle uncapping, pill extraction, syringe dispensing — without task-specific training. [source]
- Tesla Optimus V3 entered internal factory deployment in Jan 2026 with ~1,000 robots in production at Fremont and Gigafactory Texas; Tesla targets mass production at 1M units/year by Q3 2026. [source]
- 1X NEO took 10,000 pre-orders in 5 days (Oct 2025); the Hayward, CA factory began full-scale production in Q1 2026 with consumer delivery targeted for late 2026. [source]
- Unitree G1 launched at $16K (April 2026) with 23 DOF; Unitree filed for a Shanghai IPO at ~$610M (March 2026), aiming for 20K shipments in 2026 (vs ~5.5K in 2025). [source]
- Boston Dynamics' electric Atlas entered commercial production in Jan 2026 with all 2026 units committed to Hyundai and Google DeepMind, supported by autonomous battery-swapping for continuous operation. [source]
Key people
- Sergey Levine Co-founder & Research Lead · Physical Intelligence [source]
- Jitendra Malik Professor, Robotics Researcher · UC Berkeley EECS [source]
- Dieter Fox Senior Director, Robotics Research · NVIDIA / Allen Institute for AI [source]
- Pieter Abbeel Robotics & RL researcher · Amazon / UC Berkeley [source]
- Brett Adcock CEO & Founder · Figure AI [source]
- Marc Raibert Founder · Boston Dynamics; Boston Dynamics AI Institute [source]
Startups & labs to watch
- Figure AI Figure AI · STARTUP · Series C ~$1.5B at $39.5B valuation (Sep 2025); ~$1.9B total raised — Now fully independent from OpenAI on its own Helix VLA stack; reported $39.5B valuation (Sep 2025) and BMW production deal — a likely IPO candidate. [source]
- Unitree Robotics Unitree Robotics · STARTUP · Series C ~$520M at $5.5B (Feb 2026); IPO ~$610M filing — Best price/performance leader globally (G1 at $16K); Shanghai IPO filing March 2026 unlocks ~$610M for 4× capacity expansion targeting 20K units in 2026. [source]
- Physical Intelligence Physical Intelligence (Pi) · STARTUP · Founded 2024 by Levine, Abbeel et al; venture-backed (specifics undisclosed) — The leading VLA foundation-model lab; π0.5 (April 2026) shows emergent open-world generalization, positioned to license to OEMs across the industry. [source]
- Sanctuary AI Sanctuary AI · STARTUP · Series B ~$50–100M (estimate); Magna strategic partnership — 75-DOF hydraulic hands with industry-leading dexterity and a Magna automotive partnership for production deployment. [source]
- Figure 03 production scale-up Figure AI · STARTUP · Series C+; ~$39B reported valuation — First humanoid OEM to credibly hit hourly production cadence with autonomous (non-teleop) locomotion on uneven terrain - a leading indicator of commercial humanoid roll-out. [source]
- ROBOTERA ROBOTERA (Beijing) · STARTUP · >$200M (May 8, 2026) led by SF Group, HSG, IDG Capital — China's leading commercial humanoid; logistics product-market fit with China Post and SF Group; >95% of core components developed in-house [source]
- Unitree Robotics Unitree Robotics · STARTUP · Targeting ~$6.2B valuation, raising ~4.2B yuan (~$616M) via STAR Market IPO; backers include Meituan, Sequoia China, Tencent, Alibaba, Ant Group — First embodied-AI company cleared for a China A-share (STAR Market) IPO and chosen as the hardware base for NVIDIA's open Isaac GR00T Reference Humanoid; 2025 revenue 1.699B yuan at ~60% gross margins, but carries U.S. national-security/legislative risk. [source]
- NEURA Robotics NEURA Robotics GmbH · STARTUP · Series C of up to $1.4B (led by Tether; Qualcomm, Amazon, NVIDIA, Bosch, Schaeffler, EIB) — One of the largest robotics/physical-AI rounds on record; positions a European player against US/China humanoid leaders with a shared-learning 'Neuraverse' and 4NE1 humanoid. [source]
- Theker Theker · STARTUP · $85M Series A (led by CRV; Samsung, Aglaé Ventures) — Reconfigurable-form factory robots as an alternative to fixed humanoids; strategic backing from Samsung, LVMH's Aglaé, and Inditex signals manufacturing/retail pull. [source]