Robotics & Humanoid Automation
Humanoid robotics, sim-to-real transfer, embodied AI, manipulation and locomotion research
Humanoid Locomotion and Manipulation: Current Progress and Challenges
Comprehensive survey of humanoid loco-manipulation: control, planning, RL, foundation models, tactile sensing
Humanoid Agent via Embodied Chain-of-Action Reasoning for Zero-Shot Loco-Manipulation
First zero-shot loco-manipulation framework using foundation models on physical humanoids
Compliant In-Hand Rolling Manipulation Using Tactile Feedback
Compliant in-hand rolling manipulation using tactile feedback on Allegro hand with Visiflex fingertips. Forward/inverse mechanics + vision-tactile feedback controller.
Text2Touch: LLM-Designed Reward Functions for Tactile In-Hand Rotation
LLMs autonomously design reward functions for in-hand rotation policies on real tactile Allegro Hand with TacTip sensors
Humanoid Robots and Humanoid AI: Review, Perspectives and Directions
Proposes a three-paradigm framework (human-looking, human-like, human-level) and introduces the 'humanoid humanity dilemma' concept, advocating for 'humane humanoids' that integrate GenAI and human-level cognition rather than mere physical resemblance
Advancements in Humanoid Robots: A Comprehensive Review and Future Prospects
Systematic review of humanoid robot structural design, control mechanisms, and perception systems, proposing biomimetic and brain-inspired computing integration as the primary pathway for next-generation humanoid development
Safe Human-to-Humanoid Motion Imitation Using Control Barrier Functions
Vision-based humanoid imitation with CBF-QP safety filter from a single camera.
Ψ₀: An Open Foundation Model Towards Universal Humanoid Loco-Manipulation
Two-stage training (800h human video + 30h robot data) outperforms baselines with 10× more data
π₀.₅: A Vision-Language-Action Model with Open-World Generalization
Co-training across heterogeneous sources (multi-robot + web + semantic prediction) for open-world VLA generalization
Humanoid World Models: Open World Foundation Models for Humanoid Robotics
Extends world-models research from tabletop/AV to humanoid form factors
DexSim2Real: Foundation Model-Guided Sim-to-Real Transfer for Generalizable Dexterous Manipulation
VLM-as-realism-critic domain randomization (FM-DR) + tactile-visual cross-attention policy (TVCAP) + progressive skill curriculum (PSC). 78.2% real-world success across six contact-rich tasks; sim-to-real gap cut to 8.3%; beats DrEureka and DeXtreme
The Sim-to-Real Gap of Foundation Model Agents: A Unified MDP Perspective
Reframes foundation-model-agent robustness as a classical sim-to-real problem over the four MDP elements (Observation/Action/Transition/Reward); argues the LLM-agent field should import robotics' domain randomization rather than reinvent it; proposes unified vocabulary + standardized stress-test benchmarks
S²-VLA: State-Space Guided Vision-Language-Action Models for Long-Horizon Manipulation
State-Space Guided Adaptive Attention (SSGAA) tracks task progression to dynamically gate visual/language/action fusion; 2B-parameter model beats 7B-scale baselines on LIBERO/SimplerEnv long-horizon manipulation
Efficient Sim-to-Real Transfer of World-Action Models from Synthetic Priors
First successful sim-to-real transfer of a world-action model (Cosmos Policy video-diffusion) for manipulation, trained on ~800 synthetic demos/task with zero real data; 35% zero-shot success on a physical Franka arm
ROSA: A Robotics Foundation Model Serving System for Robot Factories
Shared GPU-pool serving for robot fleets (replacing per-robot edge-GPU assumption) + robotics-aware multi-model pipelines + factory-objective scheduling; up to 12.06x factory-productivity improvement over dedicated serving
ACE-Brain-0.5: A Unified Embodied Foundational Model for Physical Agentic AI
Extends ACE-Brain-0 into a closed-loop 8B-parameter unified embodied foundation model (spatial perception, decision-making, embodied interaction, self-monitoring, self-improvement) via new SSR+ training recipe; improves 14/18 spatial perception benchmarks
Xiaomi-Robotics-1: Scaling Vision-Language-Action Models with over 100K Hours of Real-World Trajectories
Scales VLA pre-training to 100k+ hours of real-world UMI-collected trajectories with auto-labeled language conditioning; unsaturated scaling; new SOTA on RoboCasa365 (57.4% vs 46.6%) and RoboDojo (20.07 vs 13.07)
1X Releases World Model for NEO Humanoid Robot
World model for autonomous learning; NEO preorders at $20K; Q2 2026 delivery
ABB + NVIDIA RobotStudio HyperReality: 99% Sim-to-Real Correlation
99% sim-to-real correlation; 0.5mm positioning; identical firmware virtual/physical controllers
Figure 03 + Helix 02: Household Tasks from Mocap and Simulation
Household task execution from mocap + simulation; Figure 03 White House demo
Boston Dynamics Electric Atlas: Production-Ready Humanoid
Electric Atlas CES 2026. 56 DOF, 50kg lift, 2.3m reach. $150K estimate. Shipping to Hyundai + DeepMind. Auto battery swap. 30K/yr factory planned 2028.
Tesla Optimus Gen 3: Factory Deployment at Scale
1000+ Optimus Gen 3 deployed in Tesla factories. Autonomous parts processing. 50K-100K target 2026. Fremont repurposed Q2. 10M/yr factory under construction.
GR00T N1.7: NVIDIA Open Foundation Model for Generalist Humanoid Robots
NVIDIA's open commercial humanoid VLA — 20K hours of EgoScale human video pretraining with Isaac Sim integration
From Vision-Language Models to Robot Control — Without Forgetting (Princeton)
VLM2VLA pipeline converts VLMs into VLA policies without catastrophic forgetting — natural-language action representation
TrendForce: China Humanoid Output to Surge 94% in 2026; Unitree and AgiBot to Capture ~80% Share
Global humanoid shipments forecast to breach 50,000 units in 2026 (~+700% YoY); China output +94%; Unitree + AgiBot ~80% of Chinese shipments. Unitree commits 75K humanoid + 115K quadruped/yr capacity; AgiBot scaled 1,000→10,000 units
Figure 03 Ramps at BotQ (1 robot/hour) and Deploys Commercially at BMW Spartanburg
BotQ output 1/day → 1/hour (24x in <120 days), 350+ Figure 03 delivered; paid commercial deployment at BMW Spartanburg (~40 units, body-shop/assembly); Helix VLA + new System 0 (S0) perception-conditioned whole-body RL controller
Tesla Optimus Cheers On Boston Marathon Runners — First US Public Debut
Optimus first US public uncontrolled-environment debut at Boston Marathon (Apr 21). Production timeline: summer 2026 low-volume, 2027 high-volume
Beijing E-Town Half-Marathon: Chinese Humanoid Robots Race 21.1km
21.1 km humanoid endurance race in Beijing (Apr 19) with Unitree, Booster, Fourier, Agibot competing
Figure 03 at $39B Valuation — BMW Spartanburg Pilot Scaling
Figure 03 at $39B valuation; BMW Spartanburg pilot scaling to 30k+ vehicle production. Horizontal humanoid platform positioning