Today on The Robot Beat: Skild AI follows its $1.4B raise by acquiring Zebra's robotics automation unit for fleet management scale, Unitree launches humanoid robots on AliExpress starting at $6,800, Tesla tapes out its AI5 custom chip, and Alibaba enters embodied robotics with benchmark-topping foundation models. Chinese robotaxi companies race to establish Middle East footholds while Wayve locks in all four major automotive chip vendors. We also cover major advances in sim-to-real transfer, tactile sensing, force sensor supply chains, and the emerging quadruped robot market in China.
Alibaba's Amap division will debut its first embodied robot — a quadruped — at the April 19 Beijing E-Town Half Marathon. Its ABot-World embodied world model has topped two international benchmarks, outperforming Google and NVIDIA models. ABot-M0 (manipulation) and ABot-N0 (navigation) also achieved state-of-the-art results, with ABot-M0 released open-source.
Why it matters
A Chinese tech giant with vast compute, supply chain relationships, and a massive mapping platform (Amap/AutoNavi) now has benchmark-leading robot foundation models and physical hardware to deploy them simultaneously. The open-source ABot-M0 release could accelerate China's broader ecosystem — smaller companies get a free manipulation model baseline. Combined with AGIBOT's AgiQuad spinout (see below), China's quadruped market is suddenly crowded with well-funded entrants on the same day.
Alibaba's combination of mapping data, cloud infrastructure, and e-commerce logistics networks provides unique training and deployment advantages beyond what a pure robotics startup can match. Benchmark leadership doesn't guarantee real-world reliability, but the simultaneous hardware launch signals commitment beyond paper publishing.
Unitree Robotics began selling its R1 humanoid internationally on AliExpress: the R1 AIR at ~$6,800 and the full R1 at $8,150 including import costs. The 1.2-meter, 29kg robot has 26 DOF with dynamic capabilities including cartwheels and downhill running. Unitree sold 5,500 humanoids in 2025 with 300%+ revenue growth and is preparing a 4.2B yuan Star Market IPO. Deliveries expected late June 2026.
Why it matters
This undercuts the $4,800 Futuring Robot F2 (covered yesterday) on mobility breadth, and dramatically undercuts Chery's Aimoga at $41,830 — but the distribution model is the real shift. AliExpress bypasses traditional robotics sales channels entirely, potentially bootstrapping a developer ecosystem ahead of Unitree's IPO. The e-commerce channel is architecturally new for humanoid robots, not just a price story.
The IPO timing link — demonstrating global commercial traction to Star Market investors — adds a strategic layer beyond ecosystem development. Whether AliExpress buyers can develop meaningful applications remains the key skeptic question, distinct from the pricing debate around Futuring Robot F2 covered yesterday.
Developed at the University of Electronic Science and Technology of China, Kangbao is a 1.2-meter companion healthcare robot beginning trials in Chengdu elderly care facilities. It measures heart rate, blood pressure, and body temperature without physical contact via infrared sensing and microvascular vibration analysis, achieving 97% physiological accuracy and 92% emotion recognition across 13 emotional states. The robot monitors entire floors, maintains medication records, and issues proactive health alerts — all with on-device processing for privacy protection.
Why it matters
Non-contact health monitoring that actually works at clinical-grade accuracy (97%) represents a significant capability for assistive robotics in elder care. Unlike companion robots that primarily provide social interaction, Kangbao combines continuous physiological monitoring with emotional awareness — creating a system that could detect health deterioration before human caregivers notice. The deployment in real care facilities (not labs) provides meaningful validation. China's aging population crisis — 400 million people over 60 by 2035 — makes this a genuine market need, not a solution looking for a problem.
The 97% accuracy claim needs validation against clinical gold standards across diverse populations, but if it holds, this could reduce the caregiver burden in understaffed facilities. The on-device processing architecture avoids privacy concerns that have plagued connected health devices. However, emotional recognition accuracy (92% across 13 states) likely drops significantly in real-world conditions with varied lighting, cultural expression differences, and medical conditions that alter facial expressions.
Greenworks introduced the AiMowbot C Series across Europe and the UK, a wire-free AI-powered robotic lawn mower eliminating perimeter wire installation. The mower uses RTKVision 2.0 navigation achieving 2.54cm accuracy, automatic obstacle detection for 200+ object types, front-wheel drive with oversized wheels for slopes, and SmartCut electrically adjustable cutting heights from 20mm to 90mm.
Why it matters
Perimeter wire installation has been the single biggest barrier to robotic mower adoption — it's expensive, time-consuming, and requires professional installation. Wire-free RTK-based navigation at 2.54cm precision eliminates this barrier entirely, making robotic mowing as simple as unboxing and mapping your yard. The European launch is strategically important because Europe is the largest robotic mower market globally, dominated by Husqvarna's wire-based systems. The 200+ obstacle type recognition suggests significant investment in computer vision training for outdoor environments.
The competitive landscape is intensifying: Husqvarna's 308V offers wire-free but relies on Wi-Fi connectivity (as highlighted in a separate review today); Greenworks bets on RTK for infrastructure independence. RTK positioning requires satellite signal quality and may struggle under dense tree cover. The broader trend is clear: the next generation of consumer robotics eliminates installation complexity to achieve mass-market adoption.
New today beyond yesterday's Boston Dynamics/Gemini Robotics-ER 1.6 industrial integration story: Spot autonomously completed household chores from a handwritten to-do list — arranging shoes, disposing of trash, gathering laundry, checking under furniture, and walking a dog. Separately, instrument reading accuracy is now reported at 98% (up from the 93% figure in yesterday's briefing), compared to 23% in prior model versions.
Why it matters
The 93%→98% accuracy update is a meaningful correction to yesterday's figure. More substantively, the household demo extends Spot's demonstrated autonomy from industrial inspection into unstructured residential environments — a domain where quadrupeds haven't previously shown reliable multi-step task execution. The jump from 23% to 98% on instrument reading in one model generation quantifies the improvement pace.
Cadence Design Systems and NVIDIA announced an expanded partnership to integrate Cadence's high-fidelity physics simulation engines with NVIDIA's Isaac and Cosmos AI training platforms. The collaboration aims to generate more accurate training data for robot AI systems by improving physics simulation fidelity, directly targeting the sim-to-real gap that remains one of robotics' most persistent technical challenges.
Why it matters
The sim-to-real gap is the bottleneck that separates impressive simulation results from reliable real-world robot behavior. Cadence brings decades of physics simulation expertise from electronic design automation — where simulation accuracy directly determines whether chips work — into robotics training. If Cadence's simulation engines can meaningfully improve the physical accuracy of NVIDIA Isaac environments, this could reduce the cost and iteration time for training embodied AI policies by orders of magnitude. Watch for benchmarks comparing policy transfer rates using Cadence-enhanced simulations versus standard Isaac environments.
The partnership is logical: NVIDIA has the robotics training ecosystem (Isaac Sim, Cosmos) but relies on approximate physics models; Cadence has gold-standard physics simulation but no robotics presence. Skeptics will note that more accurate simulation doesn't automatically solve the sim-to-real gap — domain randomization and real-world fine-tuning remain necessary. But if fidelity improvements reduce required real-world data by even 2-3x, the economics of robot training change substantially.
Toyota Research Institute demonstrated that large behavior models pretrained on diverse robot datasets achieve comparable manipulation performance with 3-5x less task-specific training data. The study used 1,700 hours of demonstrations across 500+ tasks, validated through 1,800 real-world trials.
Why it matters
This is the most rigorous empirical validation yet of the robot foundation model hypothesis — a thread running through Skild's $14B valuation, PsiBot's open-source VLA release, and Boston Dynamics' Gemini integration that we've tracked for weeks. Toyota's scale (1,700 hours, 500+ tasks, 1,800 trials) provides statistical weight that smaller studies couldn't. The 3-5x data efficiency gain directly prices the advantage: a task requiring 100 hours of demos now needs 20-30. The key open question is whether these gains transfer from tabletop manipulation to whole-body humanoid tasks.
UltraSense Systems announced an ultrasound-based tactile sensing platform that embeds sensing elements beneath the contact surface, addressing the durability limitations that have plagued traditional tactile sensors in contact-intensive robotic applications. The sub-surface architecture protects the sensor from wear and degradation while maintaining real-world touch perception. Customer evaluation kits ship June 1, 2026, targeting humanoid hands, industrial grippers, and manipulation-heavy end effectors.
Why it matters
Touch sensing is the Achilles' heel of manipulation robotics. Current capacitive and resistive tactile sensors degrade rapidly under repeated contact — a problem that becomes acute in production environments where robots handle thousands of objects daily. UltraSense's sub-surface ultrasound approach is architecturally differentiated: by sensing through the contact material rather than on its surface, the sensing element is inherently protected. If the approach delivers on its durability promise while maintaining adequate spatial resolution and sensitivity, it could unlock longer-duration autonomous manipulation across humanoid, industrial, and consumer robotics applications.
The June evaluation kit timeline suggests the technology is real but pre-production. Key questions remain: spatial resolution compared to existing tactile arrays, latency characteristics for closed-loop control, and cost per sensing unit at scale. The fact that both UltraSense (ultrasound) and Link-Touch (force/torque sensors) are attracting investment simultaneously suggests the market recognizes tactile sensing as an underserved but critical robotics component.
Link-Touch, a Chinese force sensor manufacturer, completed a Series C+ of 100M+ RMB ($14.6M) from CATL, AGIBOT, and Galbot. The company holds ~62% market share in China's humanoid force sensor market, supplying Xiaomi, XPeng, and UBTech, and plans to expand production capacity.
Why it matters
AGIBOT — whose G2 humanoids logged a verified 99.9% success rate in manufacturing this week — is now directly funding its own critical-component supplier, a vertical integration pattern with compounding effects as it scales toward 100-unit Q3 deployment. CATL's participation signals it views humanoids as a major battery customer category and is investing across the value chain. For non-Chinese humanoid developers, Link-Touch's 62% share is a supply chain concentration risk that deserves immediate attention given the dexterous manipulation benchmarks now being hit.
The investor composition tells a clearer story than the funding amount: OEMs funding suppliers to lock in allocation mirrors early automotive consolidation. Western humanoid developers should be evaluating alternative force sensor suppliers or in-house development now, before scale-up competition for supply intensifies.
Vitestro published a multicenter clinical trial in Clinical Chemistry demonstrating its Aletta autonomous phlebotomy robot achieved a 94.5% first-stick success rate across 1,633 patients, with a 0.3% hemolysis rate and 0.6% adverse event rate — both lower than manual phlebotomy benchmarks. Patient acceptance was strong: 90% reported equal or less pain than human draws, and 82% expressed preference for the robotic system.
Why it matters
This is the first peer-reviewed, multicenter clinical validation of a fully autonomous robotic blood draw system. The 94.5% first-stick rate competes with experienced phlebotomists, while the lower hemolysis (sample damage) and adverse event rates suggest the robot's consistency may actually exceed human performance. Publication in Clinical Chemistry — a top-tier diagnostic journal — provides the clinical evidence needed for regulatory approval and hospital adoption. Blood draws are among the most common medical procedures globally (~7 billion annually), making this a massive addressable market.
The 5.5% miss rate on first stick will be scrutinized — for patients with difficult veins, the failure mode and recovery protocol are critical. However, the 1,633-patient multicenter design provides statistical power that prior single-site studies lacked. Healthcare systems facing severe phlebotomist shortages may prioritize adoption even with imperfect accuracy, particularly for routine draws where the robot's consistency and availability provide throughput advantages.
Fresh off its $1.4B Series C at $14B valuation (covered yesterday), Skild AI has acquired Zebra Technologies' robotics automation business, gaining 180 engineers, fleet management software controlling 50,000+ robots, and established enterprise relationships in logistics and manufacturing.
Why it matters
Yesterday's Skild funding story established it as the highest-valued pure-play robot AI company. This acquisition answers the 'so what' question: Skild now pairs its cross-embodiment foundation model with operational deployment infrastructure at scale — a combination no other pure-play AI robotics company has. The 50,000-robot fleet is an immediate distribution channel, not a years-away buildout.
The cloud-computing parallel — platform companies acquiring deployment expertise rather than building it — is the most useful frame. Competitors like Physical Intelligence and Covariant now face a combined model+fleet advantage they'll need to address. The cultural integration challenge (enterprise software org into AI research company) is the main execution risk not previously surfaced.
AGIBOT — whose G2 humanoids completed a verified manufacturing shift at 99.9% success this week — has spun out its quadruped unit into a new subsidiary, AgiQuad, targeting $73M revenue in 2026 and 10B yuan by 2030. Mid-sized quadrupeds are reporting complete sell-outs. Alibaba's Amap is also entering the quadruped market simultaneously with its own launch.
Why it matters
The spinout separates quadruped market dynamics from the humanoid story: quadrupeds are generating near-term revenue that doesn't require waiting for humanoid scale. AGIBOT can raise dedicated capital and attract locomotion-focused talent without diluting the humanoid program. The sell-out status combined with Alibaba's entry and Unitree's AliExpress launch on the same day means China's quadruped market is entering a competitive growth phase that may generate more 2026 revenue than humanoids.
Li Liyun, who led Xpeng's autonomous driving division, is joining Chinese humanoid startup EngineAI Robotics, which closed a $200M Series B at a $1.47B valuation. EngineAI plans dozens of unit deployments in 2026, focusing on physical capabilities and motion control.
Why it matters
The AV-to-robotics talent migration thread (tracked since the Physical AI Talent War story two days ago) now has its most senior data point yet — a C-suite executive, not a mid-level engineer. Li's move from Xpeng's perception and planning stack into humanoid motion control is the clearest signal yet that China's best robotics talent increasingly favors humanoid startups over AV companies. EngineAI's $1.47B valuation places it among China's best-funded humanoid companies, giving Li resources to match the ambition.
Whether AV domain expertise — optimized for large-scale outdoor navigation — transfers effectively to the fine-grained manipulation and locomotion challenges of humanoids is the key open question. EngineAI's deliberate 'physical capabilities' focus suggests it's betting the answer is partial but meaningful.
Accenture Ventures invested in General Robotics, whose GRID platform provides unified AI orchestration across 40+ robots from different manufacturers including FANUC, Flexiv, and Ghost Robotics. The platform deploys modular AI skills across heterogeneous hardware through cloud-based orchestration and simulation-based training, addressing the fragmentation problem of factories running incompatible robot software stacks from multiple vendors.
Why it matters
Factory robot fragmentation is one of the most underappreciated barriers to scaled automation. Most facilities run robots from 3-5 different manufacturers, each with proprietary software. General Robotics' GRID platform attacks this with a vendor-agnostic AI abstraction layer — essentially middleware that lets manufacturers deploy the same AI skills across any robot. Accenture's investment signals that this orchestration layer is becoming critical enterprise infrastructure, not just a nice-to-have. For robotics entrepreneurs, this validates the platform play: building the integration layer may capture more value than building individual robots.
The parallel with enterprise IT middleware is instructive — companies like MuleSoft and Informatica built multi-billion-dollar businesses by solving integration complexity. In robotics, the middleware opportunity may be even larger because the coordination requirements are more complex (real-time control, safety constraints, physical environment variability). However, General Robotics must demonstrate that its abstraction layer doesn't sacrifice robot-specific capabilities for generality.
Korean e-commerce giant Coupang has invested $84 million (120 billion won) in AI and robotics startups over three years, with key investment Contoro Robotics developing AI-powered robotic arms achieving 99% success rates on warehouse unloading tasks. Contoro's systems use large language models to learn new tasks from natural language commands and perform self-diagnostics. Coupang is piloting the technology in Korean fulfillment centers with plans for global deployment, and has committed an additional 75 billion won to Korea's Sovereign AI Fund.
Why it matters
Coupang's investment pattern reveals how major e-commerce logistics operators are directly funding the robotics capabilities they need. Contoro's natural language command interface for industrial robots — using LLMs to learn new tasks — represents the convergence of foundation models with physical manipulation that has been theoretical until now. The 99% unloading success rate and self-diagnostic capabilities suggest production-ready reliability. The broader $84M portfolio indicates systematic robotics capability building, not one-off experimentation.
Contoro's LLM-based task learning approach is particularly interesting because warehouse unloading tasks are highly variable — mixed SKUs, random box orientations, damaged packaging. If LLMs can genuinely enable robots to adapt to new situations through natural language instruction rather than reprogramming, this could dramatically accelerate deployment across logistics operations globally.
Tesla completed tape-out of its AI5 custom inference chip: estimated 8-10x compute improvement over AI4, ~9x more memory capacity, ~5x improved memory bandwidth, consolidating GPU, ISP, and neural processing into a unified SoC. Manufacturing targets TSMC and Samsung at 2-3nm, with production expected late 2026 to early 2027. Musk confirmed AI6 and Dojo3 are in parallel development.
Why it matters
For the Tesla Optimus program — which this week confirmed Shanghai Gigafactory as a second simultaneous production hub targeting 1M units annually — AI5 is the on-device compute that determines what models Optimus can run without cloud latency. The 8-10x compute claim is significant for that use case specifically, though volume production is 12-18 months out. The parallel AI6/Dojo3 announcement suggests Tesla is accelerating its hardware cadence to keep pace with rapid model improvement.
Tesla has a history of silicon timeline misses (AI4 delayed ~6 months), and tape-out is the beginning of verification, qualification, and yield optimization — not the end. The consolidation into a single SoC mirrors Apple's M-series strategy and could reduce Optimus BOM cost. The key competitive question is whether AI5 sufficiently differentiates from NVIDIA's Jetson ecosystem and the Samsung 2nm-bound DEEPX DX-M2 covered yesterday.
Verified across 2 sources:
EVwire(Apr 15) · Electrek(Apr 15)
Artilux announced Inception, a hybrid optoelectronic AI computing architecture that combines GaN micro-LEDs and GeSi photodetectors to perform matrix multiplication using light rather than electrons. The company claims orders-of-magnitude improvements in power and area efficiency versus conventional transistor-based approaches, with a first-generation processing core under development targeting over 12,000 TOPS (INT8) at ~50 mW/mm² power density — without requiring advanced semiconductor process nodes.
Why it matters
If validated, this is a fundamental architectural departure from the transistor scaling paradigm that has dominated AI compute. By performing the most energy-intensive operation in neural networks (matrix multiplication) optically, Artilux claims to sidestep the power and thermal walls limiting conventional chips. The 12,000 TOPS target at 50 mW/mm² would be transformative for edge robotics, where power budgets are severe and thermal management is a physical constraint. The process-node independence is equally significant — if competitive performance doesn't require 2-3nm fabrication, it dramatically reduces manufacturing cost and supply chain risk.
Optical computing for AI has been promised before by companies like Lightmatter and Luminous Computing, with mixed real-world results. Artilux's GaN micro-LED approach is architecturally distinct from prior waveguide-based attempts, but the technology is pre-silicon and claims are unvalidated by independent benchmarks. The 'under development' status means production deployment is likely 3+ years away. Still, the power efficiency claims are extraordinary enough to warrant tracking — edge robotics desperately needs compute architectures that break the power-performance curve.
Wayve secured $60M from AMD, Arm, and Qualcomm Ventures, extending its Series D to $1.2B at an $8.6B valuation. Combined with existing NVIDIA backing, Wayve now has investment from all four major automotive compute platform vendors. Commercial deals include Nissan integration starting 2027 and planned robotaxi pilots with Uber in London and Tokyo.
Why it matters
Having all four major automotive chip companies as investors is unprecedented and gives Wayve's end-to-end neural network genuine hardware-agnostic positioning — a direct competitive advantage in an industry where Waymo is NVIDIA-specific and Tesla runs proprietary silicon. The Nissan 2027 integration provides the first concrete production deployment date for Wayve's approach. For the autonomous vehicle landscape this week, Wayve's model sits in sharp contrast to Uber's reported $10B+ fleet ownership commitment: one bet on platform software, one bet on operational control.
Maintaining optimization across four silicon architectures simultaneously is a real engineering challenge that grows with each platform added. The Uber London robotaxi pilot alignment with Waymo's London testing (reported yesterday) creates an interesting competitive dynamic in the same geography.
At least three major Chinese AV companies are accelerating Middle East expansion: WeRide has launched fully driverless, fare-charging robotaxi service in Dubai; Didi plans UAE testing later this year; and Baidu's Apollo Go has begun Dubai operations targeting 1,000+ vehicles.
Why it matters
The Middle East is emerging as the de facto proving ground for Chinese AV technology — offering regulatory speed, high-value markets, and freedom from U.S.-China tech restrictions. This creates a compounding dynamic: while Waymo pursues London (covered yesterday) under established regulatory frameworks, Chinese AV companies are accumulating real-world operational data and revenue in permissive markets. Baidu's 1,000+ vehicle Dubai ambition would make it one of the largest robotaxi deployments globally, rivaling Waymo's San Francisco footprint.
Operating conditions in Dubai — extreme heat, sandstorms, unique driving culture — create specific technical challenges absent from Chinese and U.S. test environments. The competitive risk for Western AV companies is less about technology gaps and more about being outpaced commercially in key growth markets before regulatory approvals align.
Lyft's Flexdrive subsidiary is constructing an 80,000-square-foot autonomous vehicle depot in Nashville to service and maintain Waymo's robotaxi fleet beginning this fall, hiring 70+ technicians, with Flexdrive already scouting European locations for future expansion.
Why it matters
This crystallizes a division of labor emerging across this week's robotaxi coverage: Waymo builds the self-driving system, Uber commits $10B+ to fleet ownership and customer infrastructure, and now Lyft's Flexdrive transforms into the physical maintenance backbone. The European depot scouting aligns directly with Waymo's London testing timeline covered yesterday, suggesting a coordinated transatlantic rollout rather than opportunistic expansion.
The depot model creates operational lock-in for Lyft but also makes it dependent on Waymo's deployment pace and geographic choices. Once you're managing Waymo's fleet in a city, switching costs run both ways.
China's Robotics Ecosystem Is Vertically Integrating at Speed Multiple stories today — AGIBOT spinning out AgiQuad, Link-Touch's force sensor dominance with CATL backing, Alibaba entering embodied robotics, and Unitree's AliExpress global sales — reveal China's robotics industry rapidly consolidating across hardware, AI, sensors, and distribution. OEMs are investing in their own suppliers, tech giants are building foundation models, and consumer channels are going direct-to-global. This vertical integration pattern is outpacing the more fragmented Western approach.
Sim-to-Real and Data Efficiency Are the New Competitive Frontiers Cadence-NVIDIA's simulation partnership, Toyota Research's 3-5x data efficiency with large behavior models, Sim2Real-VLA's zero-shot transfer, and the scalable trajectory generation work all converge on the same insight: reducing the cost and time to train robots for real-world deployment is now the critical bottleneck. Companies that solve data efficiency will deploy faster and cheaper.
Custom Silicon Strategies Are Diverging Across the Stack Tesla's AI5 tape-out for edge inference in vehicles and robots, Meta's multi-gigawatt MTIA commitment with Broadcom for data center inference, and Artilux's optoelectronic architecture represent three distinct bets on custom AI hardware. The common thread: general-purpose GPUs are insufficient for production-scale AI workloads, and the winners will be those who match silicon architecture to specific deployment constraints.
Fleet Infrastructure Beats Algorithms in Autonomous Mobility Across robotaxis, autonomous trucks, and warehouse robots, today's stories emphasize operational infrastructure over software breakthroughs. Lyft builds Waymo maintenance depots, CaoCao vertically integrates vehicle-to-fleet operations, Volvo emphasizes industrial-grade manufacturing discipline, and Skild AI acquires fleet management software. The message: scaling autonomous systems requires logistics, maintenance, and operational excellence as much as better AI.
Tactile Sensing and Force Feedback Emerge as Critical Differentiators UltraSense's ultrasound tactile platform and Link-Touch's 62% market share in humanoid force sensors highlight that manipulation capability — not just locomotion or vision — is becoming the key differentiator for next-generation robots. As humanoids move into assembly and household tasks, the quality and durability of touch sensing directly determines task success rates.
What to Expect
2026-04-19—Beijing Humanoid Robot Half-Marathon — 100+ teams including Alibaba's Amap debut of its first embodied quadruped robot, testing autonomous endurance and locomotion in public.
2026-04-21—HANNOVER MESSE 2026 opens — over 3,000 exhibitors showcasing industrial AI and robotics, with ~15 companies demonstrating humanoid robots including Agile Robots and Siemens.
2026-05-15—Public comment deadline for Aurora Operations' 5-year exemption request from federal warning device regulations for autonomous trucks.
2026-06-01—UltraSense Systems begins shipping ultrasound tactile sensing evaluation kits for robotics developers.
2026-06-30—Unitree R1 humanoid robots expected to begin deliveries to international AliExpress customers.
How We Built This Briefing
Every story, researched.
Every story verified across multiple sources before publication.
🔍
Scanned
Across multiple search engines and news databases
844
📖
Read in full
Every article opened, read, and evaluated
174
⭐
Published today
Ranked by importance and verified across sources
20
— The Robot Beat
🎙 Listen as a podcast
Subscribe in your favorite podcast app to get each new briefing delivered automatically as audio.
Apple Podcasts
Library tab → ••• menu → Follow a Show by URL → paste