Skip to content

Radar — Queue / Watchlist

Last updated: 2026-08 · owner: Youngjin · volatility: high ← back to index

L0 TL;DR: Things worth watching that have not yet passed the inclusion criteria (2.5 THE FILTER). Each item is one line — a maturity label + why it's noteworthy + why it is on hold. Once it clears the gate (2 of 4), the owning pillar's owner promotes it using the standard template.

⚠️ Do not present items here as "mature capabilities" in customer proposals. A flashy demo often masks how deployable something actually is.

🔬 Models / algorithms (awaiting validation)

Item Label At a glance Promotion condition
Physical Intelligence π0.7 🔵 Research ✨ Notable: Rumored next flagship from PI, a leading VLA lab (π0/π0.5) — a release could move the industry baseline again
⏳ On hold: Secondary sources only [4], no primary PI confirmation		 Official PI release + performance validation
GR00T N1.6 / N1.7 commercial license 🟡→ ✨ Notable: If commercial use is real, it becomes a rare open VLA usable in customer proposals (N1.5 is non-commercial, so unusable)
⏳ On hold: Commercial-use claim is secondary-source only [4] (N1.5 is clearly non-commercial per model card [1])		 License confirmed on live model card
World-action models (DreamZero → GR00T N2) 🟡 Preview ✨ Notable: The "world models that also generate actions" track touted as the post-VLA generation — a directional indicator for NVIDIA's roadmap
⏳ On hold: GR00T N2 "expected end of year," DreamZero is research		 GA + real deployment case
Google DeepMind Genie 3 (world model1 for robot learning) 🟡 Preview ✨ Notable: An attempt to use a frontier-grade world model as a data source for robot policy learning — would bypass the real-data bottleneck if it works
⏳ On hold: The world model itself is preview; applying it to robot learning is research		 Validated case of robot policy learning
VLM-based SysID2 (Vid2Sid, Swim2Real) 🔵 Research ✨ Notable: Estimates physical parameters from video alone to automate simulator calibration — could eliminate manual sim-to-real calibration
⏳ On hold: 2026 preprints, single lab		 peer-review + reproduction
VIRAL / VideoMimic / Real2Render2Real (visual sim-to-real3 at scale) 🔵 Research ✨ Notable: Visual sim-to-real that reconstructs simulation environments and demonstrations from ordinary video — a candidate to change the data-collection cost structure
⏳ On hold: CVPR/CoRL research, not production		 Evidence of production deployment
Robbyant LingBot-VLA / UnifoLM-VLA-0 🔵 Research ✨ Notable: A new wave of open VLA families from China — watched for the open-weights competitive landscape
⏳ On hold: Secondary sources, no validation		 Primary confirmation + AWS mapping

🖥️ Simulation / tools (awaiting maturity)

Item Label At a glance Promotion condition
Genesis physics engine4 ⚪ Hype ✨ Notable: Hyped as an "ultra-fast general-purpose physics engine" — if true, it changes the cost structure of GPU simulation
⏳ On hold: "430,000×" refuted [1], slow on contact-rich manipulation		 Independent benchmark + production adoption
MuJoCo Warp 🟡 Alpha ✨ Notable: Combines MuJoCo accuracy with GPU parallelism — an alternative candidate to the Isaac monoculture
⏳ On hold: PyPI classifier "3-Alpha" [1], not production		 Beta/GA transition
NVIDIA Newton physics engine 🟡 Preview ✨ Notable: Next-generation open-source physics engine co-developed with Google DeepMind and Disney Research — the likely next standard of the Isaac ecosystem
⏳ On hold: Experimental backend in Isaac Sim 6.0		 GA + official in Isaac Lab 3.0
Isaac Sim 6.0 🟡 Preview ✨ Notable: A next-generation overhaul including Newton integration — a directional indicator for migrating the current 5.x stack
⏳ On hold: "Early Developer Release," API in flux (latest GA is 5.1)		 6.x GA declaration
Cosmos 3 as sim-to-real training source 🟢 GA(model)/🔵(in practice) ✨ Notable: The track of training deployable policies from world-model-generated data — would reshape the SDG pipeline landscape if it works
⏳ On hold: Model is GA, but "training deployable policies from world-model data" is early-adopter only. ⚠️ not hosted on AWS		 Stronger AWS mapping + training validation

🤖 Hardware / deployment (roadmap / demo)

Item Label At a glance Promotion condition
Tesla Optimus V3 ⚪ Hype ✨ Notable: The highest-profile humanoid mass-production plan — the most frequently asked-about item by customers
⏳ On hold: Musk claims only, production not started		 Validated deployment
Hyundai·BD electric Atlas ⚪ Roadmap ✨ Notable: Hyundai Motor Group's mass-production roadmap (30k/yr from 2028) — the most direct humanoid track for customer conversations in Korea
⏳ On hold: Electric Atlas product version unveiled (2026-07, BD official [3]). Deployment 25k+ units & 30k/yr capacity both start 2028; ~0 in live production today. 2026 is a small pilot only (Hyundai RMAC + Google DeepMind). ⚠️ "Gen 5" is a misnomer		 Verified real-operation shipments begin
Apptronik Apollo 2 + Robot Park 🟡 Pilot ✨ Notable: Live operational pilots at Mercedes and GXO plus a Google DeepMind data partnership — a front-line indicator of humanoid commercialization
⏳ On hold: Operational pilots at Mercedes-Benz & GXO [3] + Google DeepMind Gemini Robotics data partnership (90k sq ft). Autonomy/commercial scale unverified. AWS mapping is generic (data→S3/SageMaker); the partnership itself is Google [4]		 Commercial deployment scale + validated autonomy
1X Neo autonomy 🟡 Preview ✨ Notable: Among the first home humanoids actually on sale ($20k) — a test bed for the mixed-teleoperation operating model
⏳ On hold: Mixed autonomy + VR teleoperation (Expert Mode) — acknowledged by the CEO directly (Engadget [3]). The "60–70% autonomy" figure has no primary source [4]		 Validation of true autonomy
Figure 03 "8-hour autonomous shift" ⚪ Hype ✨ Notable: An autonomy claim on top of a validated BMW pilot record — would reset the bar for industrial humanoid autonomy if true
⏳ On hold: CEO tweet, no independent validation (Figure 02@BMW is a validated pilot)		 Third-party autonomy audit
Cosmos 3 adoption (Doosan/LG/Samsung) 🟢 GA(announced) ✨ Notable: Adoption announcements by three major Korean conglomerates — a reference that comes up immediately in customer conversations in Korea
⏳ On hold: Adoption is "announced," not production-validated		 Public production case

🔗 Agents / connectivity (early)

Item Label At a glance Promotion condition
MCP5 for robotics (ros-mcp-server, etc.) 🔵 Research ✨ Notable: A surge of experiments (50+ servers) wiring the agent-standard protocol to robot skills — an AgentCore-integration angle
⏳ On hold: 50+ servers exist but open-source/demo, none in production (safety, latency, determinism unvalidated)		 Production-hardening case
ROS 26 + LLM agents7 (NASA JPL ROSA, RAI) 🔵 Research ✨ Notable: Real-organization validation cases such as NASA JPL's ROSA — the most realistic entry path from natural language to robot operations
⏳ On hold: ROSA (JPL) is the strongest real case but mock-ops. Field deployment limited		 Field production deployment
Agent physical-safety standards (RoboGuard, etc.) 🔵 Research ✨ Notable: A standards vacuum for LLM semantic-level risk — could surface as a regulatory/procurement requirement
⏳ On hold: ISO covers physical only; no standard for LLM semantic risk		 Progress on standardization
AgentCore Payments / Agent Registry (Seoul) 🟡 Preview/unavailable ✨ Notable: The AWS-native track for robot-agent commerce and registry infrastructure — immediately proposable once it opens in the Seoul region
⏳ On hold: Not available in Seoul region — Agent Registry is in Tokyo ✅, but Payments is not in Tokyo either (Sydney only in APAC) [1]		 Seoul region expansion

🆕 Latest scan intake (2026-09-05 · primary verification completed 2026-07-21)

Item Label At a glance Promotion criteria
Walden Robotics (Toyota Research Institute spinout, Large Behavior Models9 humanoid) 🟡 Pilot ✨ Notable: A spinout by Russ Tedrake, who led TRI robotics, with a $300M seed — the front line of LBM commercialization, with a live Toyota-plant pilot
⏳ On hold: Official company announcement (2026-07-15) [4] — spun out of TRI in 2026-01 (founder Russ Tedrake, former TRI SVP); $300M seed co-led by Toyota and Deviation Capital, with NVIDIA, Boeing, Samsung Ventures, etc. participating ($1.1B valuation). Humanoid upper body on a wheeled mobile base, running Diffusion Policy10 / Large Behavior Models; claims a pilot-to-"production" transition at a North American Toyota plant starting 2026-02, no third-party validation		 Third-party audit / independent validation + expanded deployment scale
Xiaomi-Robotics-1 (VLA8 foundation model, 100k+ hours of real-world UMI11 trajectories) 🔵 Research ✨ Notable: Claims SOTA on four benchmarks on the strength of 100k+ hours of real-world UMI trajectories — a signal that Chinese big tech is entering the VLA race in earnest
⏳ On hold: Xiaomi official arXiv release (2607.15330, 2026-07-16) [4] — Qwen3-VL-based MoT (VLM+DiT), self-reported SOTA on four benchmarks including RoboCasa365 (57.4%, up from prior SOTA 46.6%) (compared against RLDX-1 and GR00T N1.6 among others; no independent reproduction). ⚠️ Correction (2026-08-10): confirmed code/checkpoints (base 5B + 3 task-specific variants for RoboCasa/RoboCasa365/VLABench) released on GitHub on 2026-08-03 — Apache-2.0 (unlike RLDX-1, explicitly permits commercial use, opening up AWS-mapping potential). The repo's own leaderboard numbers use a different reporting format than the arXiv paper and need direct cross-checking; independent reproduction and real deployment remain at zero		 Independent benchmark reproduction + real deployment case
Gemini Robotics 2 (Google DeepMind, whole-body-control VLA) 🟡 Preview ✨ Notable: A frontier-lab VLA expanding beyond upper-body manipulation to whole-body control (walking, two-hand coordination) — a generational shift that reshapes the competitive-stack landscape
⏳ On hold: Official announcement (2026-07-30) [4] — extends prior upper-body-only control to whole-body control (walking, bending, two-hand coordination); launched alongside a reasoning model, Gemini Robotics ER 2, and an edge model, On-Device 2. Demoed on Apptronik Apollo 2 (92% success unscrewing a light bulb), self-reported benchmark, no independent validation. Only ER 2 has a public preview (AI Studio/Enterprise Agent Platform); VLA and On-Device 2 are early-access-partner only. ⚠️ pillar-2's "Gemini Robotics" competitor-stack section is a pre-announcement snapshot (confirmed 2026-07, covers ER 1.6/On-Device/1.5) — needs a pillar-owner refresh		 Early access ends, GA release + independent benchmark validation
XYZ Robotics DEUX (dual-arm semi-humanoid built on the "Physical AI Data Flywheel" of BrainX, GloveX, and TwinX) 🟡 Preview ✨ Notable: A Korean startup trains BrainX on real data from its own commercial café and delivery robot operations (its barista robot has processed 1M+ orders), then chains teleoperation capture via GloveX → digital-twin validation in TwinX → redeployment to real hardware — a local Physical-AI data-pipeline case that could come up directly at Korean customer touchpoints
⏳ On hold: Field trial started at LoungeX's Seongsu flagship store in Seoul (2026-07-27) [4] — dual 32-DoF arms with three-finger hands; raised a KRW 13B Series B (2026-03). Both DEUX's commercial launch (planned H2 2026) and GloveX's external sale (planned late 2026) are still unreleased. Autonomy level and deployment scale remain unvalidated		 Expanded field trials + independent validation of autonomy performance
ROBOTIS AI Sapiens K1 (open-source humanoid platform built on DYNAMIXEL-Q) 🟡 Preview ✨ Notable: An open-source humanoid from Korean company ROBOTIS — demoed learning a K-pop dance move from a smartphone video alone (video motion capture → retargeting → sim RL → sim-to-real), plus a text-to-motion extension integrated via NVIDIA Kimodo. A local open-humanoid case that comes up directly at Korean customer touchpoints
⏳ On hold: Official GitHub repo (ROBOTIS-GIT/ai_sapiens, Apache-2.0) [4] — the ROS 2 packages (robot description, controller interfaces, sim2real tools) are already released, but open-sourcing the full video-to-motion pipeline is only an announced ROBOTIS plan (no confirmed timeline). The K-pop demo is self-reported; no independent validation or real deployment case		 Full pipeline release confirmed + independent validation
AWS-NVIDIA Physical AI infrastructure expansion (Amazon Robotics × NVIDIA collaboration) ⚪ Roadmap ✨ Notable: AWS itself announced it is putting physical-AI infrastructure (simulation, SDG12, robot training, functional safety, real-to-sim validation, GPU-accelerated EC2) behind Amazon Robotics' next-generation robot development — the most directly "AWS-native" physical AI item the Radar has carried
⏳ On hold: Joint AWS/NVIDIA announcement on 2026-08-26 [4] (primary sources press.aboutamazon.com / nvidianews.nvidia.com, cross-confirmed by numerous secondary outlets) — part of a 2M-additional-GPU rollout (2027–2028). Only the scope of Jetson/Omniverse/Isaac platform use is disclosed; no specific Amazon Robotics robot, quantified result, or service name has been named — roadmap stage, 0 real deployment		 Named robot/service disclosed + confirmed AWS service mapping
LG × NVIDIA bipedal humanoid + CLOiD ⚪ Roadmap ✨ Notable: A major Korean conglomerate (LG) signed an MOU with NVIDIA — a bipedal humanoid built on Jetson Thor, Isaac GR00T, and Halos (a robotics safety framework) is planned for a Q1 2027 reveal, plus the wheeled CLOiD goes into real validation at a Tennessee washing-machine plant within 2026 — a local-conglomerate track directly citable in Korean customer conversations (a similar angle to Hyundai·BD's Atlas)
⏳ On hold: MOU signing officially announced 2026-08-13 (LG Corp Chairman/CEO Kwang Mo Koo and Jensen Huang in attendance) [4]. CLOiD's Tennessee deployment happens within 2026 but is a validation stage (not commercial scale); the bipedal humanoid has no prototype shown yet (reveal planned Q1 2027) — 0 in live operation today		 Bipedal humanoid hardware revealed + CLOiD factory-validation results published
GHOST (VR teleoperation letting one operator control two robots at once from onboard cameras alone) 🔵 Research ✨ Notable: Built by Brown University's Tellex lab with Amazon funding — a single operator teleoperates two Boston Dynamics Spot robots simultaneously in VR using only onboard RGB-D, no external motion capture, published in IEEE RA-L (peer-reviewed). Measured 1.6–4× novice success rate and a 1.47× expert speedup — an open-source case that could lower the cost of teleoperation-based robot data-collection pipelines
⏳ On hold: arXiv 2608.29080 (2026-08-29, accepted IEEE RA-L 2026-08) [4] — expert evaluators were the paper's own three authors (potential bias); novice study n=15 on only 2 of 9 tasks; hardware-specific to Boston Dynamics Spot on a dedicated Wi-Fi network (production reliability unvalidated)		 Expanded independent user evaluation + validation across varied hardware/network conditions
Perceptron Isaac 0.5 (open-weight embodied foundation model, 36B parameters) 🔵 Research ✨ Notable: Unifies video understanding, embodied reasoning, and robot control in a single sparse backbone, released as open weights — trained on 35+ robot systems, 100k+ hours of robot experience, 1M hours of video, and 3T multimodal tokens; self-reports besting π0.5 and GR00T N1.7, with code and weights both released (code Apache-2.0)
⏳ On hold: Official company announcement + official GitHub repo (2026-08-27/28, from ex-Meta-researcher startup Perceptron AI) [4] — self-reported benchmarks, no independent reproduction or peer review. The weights' own license terms are stated separately on the Hugging Face repo (access unverified here)		 Independent benchmark reproduction + real deployment case
ABEJA × Murata Manufacturing GR00T N1.7 dual-arm PoC (VLA-based physical-AI technical validation) 🟡 Preview ✨ Notable: Japanese manufacturing giant Murata Manufacturing validated a dual-arm robot on real hardware using NVIDIA GR00T N1.7 (a commercially licensed open VLA) — handing a part between arms, reorienting it, and inserting it in one continuous sequence — an early case of a real manufacturer validating a commercially licensed open VLA, from a lab-automation angle
⏳ On hold: Official ABEJA / Murata Manufacturing announcement (2026-08-31, PR TIMES) [4] — imitation learning from hundreds of teleoperated demonstrations; succeeded on real hardware in a validation (test) environment. No production-scale deployment or independent validation of autonomous performance. ⚠️ The link is the official PR TIMES release, but this run's environment had egress restrictions that blocked the manual curl-200 check (see commit message / issue)		 Production-line deployment + independent validation

⚰️ Retired — do not propose (kept for the record)

Item Status Replacement
AWS RoboMaker 🔴 Discontinued (2025-09-10) [1] EC2 G6e/G7e + Isaac Sim AMI + AWS Batch
SageMaker Edge Manager 🔴 Discontinued (2024-04-26) [1] ONNX + IoT Greengrass V2 (+ SageMaker Neo)
IoT Greengrass V1 🔴 Discontinued (2026-06-01) [1] Greengrass V2
Gazebo Classic 11 🔴 EOL (2025-01) [1] Gazebo Jetty/Harmonic
Trainium for VLA ⚪ No public case [4] Currently CUDA/NVIDIA (state the risk when proposing)

⚠️ Rumor watch (not true): "AWS IoT TwinMaker discontinued" is misinformation — TwinMaker is GA and open to new customers (low velocity). It is a third-party blog claim confused with SiteWise maintenance. Do not repeat. → pillar-3.

Promotion procedure (summary)

  1. Capture: collect candidates via a designated channel/emoji
  2. Filter: apply the 2.5 gate (2 or more of 4)
  3. If it passes: the owning pillar's owner incorporates it via the standard template and removes it from the Radar
  4. If it falls short: keep it here as a one-liner, with the promotion condition stated

Full pipeline → maintenance.

owner: Youngjin · updated: 2026-08 · volatility: high (the Radar changes fast by nature — monthly review recommended)

  1. World Foundation Model (WFM) — a large model trained to predict/generate the next scenes of the physical world. From text/video prompts it creates physically plausible video and scenarios to augment robot training data. 🎥 NVIDIA Cosmos introduction ↩

  2. SysID (System Identification) — measuring the real robot's physical parameters (friction, mass, motor response) to calibrate the simulator to the real hardware. ↩

  3. sim-to-real — transferring a policy trained in simulation to a real robot, or the methodology for doing so. The physical and visual differences between simulation and reality (the domain gap) mean a naive transfer collapses performance. 🎥 NVIDIA sim-to-real robotics showcase ↩

  4. Physics engine — the core software of a simulator that numerically computes rigid-body dynamics, contact, friction, and collision. An engine's accuracy-speed trade-off drives the choice of simulator (Isaac/MuJoCo/Genesis). ↩

  5. MCP (Model Context Protocol) — an open standard protocol connecting agents to tools and data sources. Often likened to "USB-C for agents"; experiments exposing robot skills as MCP servers are growing. ↩

  6. ROS 2 (Robot Operating System 2) — the de facto standard open-source middleware for robot software. A distributed architecture in which sensor and control nodes communicate over topics; the shared foundation of industrial and research robot stacks. ↩

  7. LLM agent — software in which a large language model plans on its own, selects and calls tools (APIs, robot skills), and carries out multi-step tasks. Unlike simple Q&A, the key point is that it "acts." ↩

  8. VLA (Vision-Language-Action) — a foundation model that takes camera images (Vision) and natural-language instructions (Language) as input and directly outputs robot actions (Action). Say "pick up the cup" and it generates the joint motions. 🎥 NVIDIA Isaac GR00T N1 introduction ↩

  9. Large Behavior Models (LBM) — the "robot behavior" counterpart of LLMs: Toyota Research Institute's term for robot foundation models trained on large demonstration data to perform many manipulation tasks with a single model. ↩

  10. Diffusion Policy — a policy architecture that generates robot action sequences with the diffusion models used in image generation. It learns demonstrations with multiple valid variations stably and has become the de facto standard for imitation learning. ↩

  11. UMI (Universal Manipulation Interface) — a data-collection method in which a human holds a camera-equipped handheld gripper, no robot required. It enables collecting real-world demonstrations at scale without deploying robots. ↩

  12. Synthetic Data Generation (SDG) — a technique that uses a simulator to auto-generate training images and annotations (labels). Its biggest advantage: labeling cost converges to zero. 🎥 Isaac Sim Replicator SDG tutorial ↩