AI Competition Radar: October 18–25, 2026
This week's scout radar: a humanoid-robotics challenge that named an electroadhesive gripper, an insurance claims system that won a buyer in Vietnam, four track winners at Google San Francisco, and two Chinese competitions that published full scored rosters.
This edition covers results published between 18 and 25 October 2026, Pacific time. Five competitions published named winners in that stretch. Three of them were run by companies rather than conference organisers, and two were in China with complete ranked rosters. Most entries this week carry a placement, an institution and a product line. Where a roster stops at names, the entry says so.
This week at a glance
| Event | Organizer and place | Result date | Top result | Source |
|---|---|---|---|---|
| Infineon Startup Challenge 2026 | Infineon; demo day at EBSCON, Graz; winners named in Munich | 19 Oct | AdapTronics takes first with an electroadhesive gripper | Robotics & Automation News |
| AI for Insurance Challenge × Tasco | GenAI Fund's BeyondPilot with Tasco; Ho Chi Minh City | 22 Oct | ALPHV Technologies wins; a paid pilot and up to $100,000 follow | BeyondPilot |
| Build for Everyone Hackathon | k-ID with Google; Google San Francisco | 21 Oct | Freetime, Ollie Ollie Out!, Curio and Growfolio take the four tracks | k-ID |
| AI + Social Work Innovation Application Competition | 2026 A-Chao league; Liuzhou, Guangxi | 22 Oct | Yiyushou wins the college track, Bianjiang Zhizhou the enterprise track | Winner list |
| Sim2Real Challenge, China Robot and AI Competition | Chinese Association of Automation and partners; Hangzhou | 18 Oct | Embodiment-SUDA wins the power-scene robotics contest | Organizer report |
Infineon Startup Challenge: a gripper wins the humanoid brief
Infineon's 2026 startup challenge asked for technology for humanoid robots and physical AI. More than 100 startups from 29 countries applied, and 15 were selected to build against Infineon's own hardware and software platforms 1.
Those platforms covered environmental sensing, artificial skin, motor control, visual feedback and Infineon's in-house AI development software. The 15 finalists presented at a demo day on 6 October, held inside the Electronic Based Systems Conference in Graz, Austria 2.
Infineon published the three winners on 19 October 3. Judges scored innovation, technical feasibility, market potential and the ability to scale 2.
AdapTronics of Italy took first place. Its gripper holds an object with electroadhesion, an electrostatic pull that works on a wide range of shapes without suction cups or mechanical fingers. Infineon points to logistics, packaging and, later, space robotics 1.
Founder and chief executive Lorenzo Agostini said the challenge gave the company access to technology experts, industry insight and an innovation network 1. As the overall winner, AdapTronics joins Infineon's Startup Co-Innovation Program, with the work due to begin at Infineon's Startup Night in Munich in November 1.
COSMOKE of South Korea placed second with COSKIN, a flexible magnetic sensing skin. The company says the material reports touch, pressure and physical contact across a large area, for dexterous manipulation and safer contact between a humanoid and the people around it 1.
TECHiFAB of Germany placed third with an in-memory computing approach to real-time motor control. The company says the design makes control inside a humanoid drivetrain faster and more energy-efficient 1.
Evidence status: the ranking, the field and the technologies come from Infineon's own release, in trade coverage on 22 October. No cash prize, no customer and no funding round is published for any of the three. First diligence question: for AdapTronics, what one gripper costs at volume, and which logistics operator has run it outside a lab.
Tasco's insurance challenge: a claims system wins, then a pilot
Tasco is a Hanoi-listed group whose businesses run along a vehicle's whole life: electronic tolling through VETC, Geely and Lynk & Co distribution, roadside assistance, charging and insurance 4. With GenAI Fund's BeyondPilot programme, Tasco's leadership set one question. How might insurance in Vietnam become something a driver is glad to hold?
The brief was compulsory motor insurance: mandated, price-regulated, sold as paper that can be lost, and claimed through a hotline. One rule barred premium discounts, cashback and disguised price incentives; every other part of the customer journey was open 4.
Nine teams were shortlisted and pitched live on 22 October, scored by GenAI Fund partners, Tasco's insurance team and judges from Microsoft, Singtel and Positive Moves 4. The nine were ALPHV Technologies, Fly Elephant Information, Frondesk Technology Solutions, Groot, Gulshan Mehta, Insider One, iorta TechNXT, Kquika and Veehive 5.
ALPHV Technologies won. The company announced the result on its own channels, saying it pitched against eight other companies from the US, Turkey, Dubai and Taiwan 6.
ALPHV's Tasco system puts photos, video and PDFs into one claim file, flags duplicate claims from the same person, and drafts a liability assessment against Vietnam's Road Act and its compensation schedule. Garage, email and WhatsApp steps run as automated workflow, and Qwen models read the images and documents 7.
The company describes the work as a proof of concept with no production results yet. Damage assessment from images and video runs around the clock; the liability assessment and claim report are drafted for a person to check 7.
The prize is a buyer rather than a cheque. Tasco confirms which teams it takes forward between 23 October and 1 November, runs a four-week pilot inside the business through November, and gives a written decision by 5 December. Teams that come through it can be considered for up to US$100,000, offered at the partners' discretion after due diligence 4.
The programme has a precedent: in September, Tasco led an investment round into BonBon, a startup it first met through its own innovation programme, with GenAI Fund investing alongside 4.
Evidence status: the organiser has published no results page, so the win rests on ALPHV's own announcement. The shortlist, the judging panel and the pilot path are the organiser's, and no team member is named on either side. First diligence question: is the Tasco proof of concept paid, and does the US$100,000 carry equity.
Build for Everyone: four track winners, no product lines
k-ID, an age-assurance vendor, and Google ran a one-day hackathon at Google's San Francisco office on 21 October. Fifteen teams of two or three took a single brief: build an app or game that becomes the right product for whoever is using it 8.
Teams built on k-ID tooling that had not been released: it resolves a user's age and jurisdiction at runtime, applies age rules across more than 200 jurisdictions, and adds parental consent or guardian controls where local law requires them 8.
Jonathan Lai of Andreessen Horowitz, Michael Carney of Google and Brian Behlendorf of the OpenSSF at the Linux Foundation judged the demos 9. Thirty-eight builders took part 9.
Freetime won Consumer AI, built by Alberto Simonato, Alessandro Castellaro and Smruti Singh. Ollie Ollie Out! won Gaming, built by Michelle Barron, Maya D. and Khushi Himanshu Gangrade. Curio won Education, built by Shehryar Hassan and Sruthi Sridhar. Growfolio won Commerce and marketplaces, built by Deepanshi and Saskia Sartor 9.
Honourable mentions went to Safelens in Consumer AI, OzoneLayer in Gaming, Reading Tools in Education and Kardza in Commerce and marketplaces. Teams named One Agent Three Generations, Homebase, Devslugs, Voicewell and Whatnow also built through the day 9.
The prizes were Google Cloud and AI credits, an expanded user allowance on k-ID, time with the judges and Google's product team, and a profile in the organiser's media push. No cash figure is published 8.
Evidence status: the roster names every member of every team, and describes nothing any of them built. k-ID says builder profiles will follow over the coming weeks. First diligence question: what Curio does differently for a nine-year-old and a fifteen-year-old, and whether any of the four has formed a company.
In Liuzhou, a full roster with scores and product lines
The 2026 AI + Social Work Innovation Application Competition held its final in Liuzhou, Guangxi, on 22 October. It is one of the topic contests inside Guangxi's A-Chao league for AI adoption across industries and public services 10.
Registration opened on 10 July and drew 1,323 teams from 15 Chinese provinces plus Laos, Malaysia and Vietnam. The top 12 in each of two tracks, one for colleges and one for companies and public institutions, reached a 24-team final, judged by nine experts over an eight-minute pitch and five minutes of questions 10.
The organisers report signed projects worth more than 20 million yuan at the event, and say winning work will be promoted into government procurement lists and social-investment funding 10.
| Track | Place | Project | Team or institution | Score |
|---|---|---|---|---|
| College | 1st | Yiyushou (易语手), AI accessibility for deaf communities in ASEAN | Liuzhou Institute of Technology | 90.28 10 |
| College | 2nd | Naoyu Kangcheng (脑驭康程), brain-computer interface for lower-limb rehabilitation after stroke | Liuzhou City Vocational College | 89.286 10 |
| College | 2nd | Zhilian Wenmai (智链文脉), Guangxi culture export to ASEAN markets | Liuzhou City Vocational College | 85.395 10 |
| College | 3rd | AI social-governance hub with a digital citizen hotline | Jinan University with the Guigang big-data and government-development bureau | 84.49 10 |
| College | 3rd | Taoxin Shouwang (陶心守望), AI for Nixing pottery kiln yields and heritage licensing | Qinzhou Preschool Education College | 84.137 10 |
| College | 3rd | Benge Zhizhi (笨格智治), AI grid services for rural areas in rocky desertification zones | Hechi University | 83.022 10 |
| Enterprise | 1st | Bianjiang Zhizhou (边江智州), city-governance decision engine | Runjian Co. (润建股份) | 85.665 10 |
| Enterprise | 2nd | Zhiban Yiyang (智伴颐养), long-term-care insurance companion and monitoring system | Liuzhou Xianyuan Health Industry with Liuzhou Luobide Technology | 85.513 10 |
| Enterprise | 2nd | AI digital social-worker cloud platform | Xunfei Zhiyuan Information Technology (讯飞智元) | 85.217 10 |
| Enterprise | 3rd | Zhuang-language translation system for the southwestern border | Baise Municipal Public Security Bureau, cyber-security detachment | 84.228 10 |
| Enterprise | 3rd | Jingxian Jinjia (敬仙进家), smart social-governance scheme for Guangxi | Guangxi Yunmeng Bagui Culture Technology | 83.83 10 |
| Enterprise | 3rd | China–ASEAN AI medical volunteer service system | Beijing Chuangxiang Tiankong Technology | 83.626 10 |
The honourable mentions carry the more product-shaped work. The college list includes a multi-agent network-service platform from Guilin Normal University, a China–ASEAN railway volunteer-matching platform from Liuzhou Railway Vocational Technical College, and a support app for autistic children from Guangxi University of Science and Technology. On the enterprise side, Liuzhou Workers' Hospital entered a fall-prevention agent for elderly patients and Liuzhou People's Hospital a digital human for medical social work 10.
One reading of the split: buyers entered the enterprise track. A municipal public-security cyber unit, two city hospitals and a social-work service centre took places next to the companies, and the roster never says which entries are incorporated startups. The college track is where a founder-shaped product first shows up, starting with the sign-language system that won it.
Evidence status: the roster, the scores and the one-line descriptions are the organisers', published two days after the final. No funding, customer or team-member detail appears for any entry. First diligence question: for Yiyushou, whether a school or provincial office has deployed it, and who holds the model and the sign-language data.
In Hangzhou, robots that had to work a live power cabinet
The first Global Embodied Intelligence Application Challenge ran with the 2026 China Robot and AI Competition in Hangzhou from 16 to 18 October. The combined event carried 22 tracks, 60 contest events and three company-run contests, with close to 5,000 competitors on site 11.
One company contest was a Sim2Real challenge for physical AI in power scenes, started by Fengyuzhu (603466.SH) and Songying Technology. Its organisers were the Chinese Association of Automation, the national embodied-intelligence application pilot base and Hangzhou Gaoxin Sci-Tech Group 11.
Forty-seven teams with more than 200 members entered. They trained robot policies in simulation on the OrcaLab platform with a Unitree G1 model, and six teams reached the on-site stage 11.
Each finalist then spent about four days at the Hangzhou base collecting data and tuning the model. On the day, each robot had to inspect a power distribution room and operate electrical cabinets on its own, with no teleoperation and no human intervention 11.
Embodiment-SUDA won, with results announced on the morning of 18 October. Zhixun Xianfeng (智巡先锋队) and HashGO were runners-up. Keke (科科), Dengdeng Fazhan (电等发展队) and AAA Wholesale Cabbage (AAA 批发卷心菜) placed third 11.
Fengyuzhu chairman Li Hui handed out the awards at the closing ceremony at the Hangzhou Baimahu International Expo Centre 11.
Evidence status: the organiser published team names only, with no member names, no institution, no prize amount and no write-up of each robot's approach. The task and the stack are the organiser's account. First diligence question: does the winning policy transfer to a substation the team has never seen, or does each site need its own four days of data collection.
Result announced, roster not yet published
- AMD Developer Hackathon: ACT III (AMD with lablab.ai; online 12–17 October, on site in Rome, Milan and Imperia on 17–18 October). The event closed with 6,083 participants on 1,373 teams and 312 projects across five tracks. Nine days on, the dashboard still reads "judging in progress", and its only ranking is by community votes, where Gara Facile of team Arepa Power leads with 343 12.
- Cal Hacks 13.0 (Hackathons at Berkeley; San Francisco, 23–25 October). The weekend closed as this edition went out, with no results posted 13.
Dates to watch
- 27 October — k-ID opens its age-adaptive tooling to the public, the same tooling the fifteen teams built on at Google San Francisco 9.
- 3 November — Monad names the Metropolis winners across four tracks, with a $25,000 grand champion inside a pool of more than $250,000. Submissions closed on 13 October 14.
- 3–8 November — the Vultr Agent Rush Hackathon builds online from 3 November, with an on-site phase in Salt Lake City on 8 November and a $14,000 prize pool 15.
- 23 October to 5 December — Tasco's path from the insurance demo day: pilot scoping until 1 November, four weeks of pilot inside the business, and a written decision by 5 December 4.
Five teams to look at first
- AdapTronics (Infineon Startup Challenge winner). Ask what one gripper costs at volume, and which logistics operator has run it outside a lab.
- ALPHV Technologies (Tasco insurance challenge winner). Ask whether the Tasco proof of concept is paid, and what the US$100,000 would buy.
- Curio (k-ID education track winner). Ask what changes in the product between a nine-year-old and a fifteen-year-old, and whether the two builders have incorporated.
- Yiyushou (Liuzhou college-track winner). Ask whether a school or provincial office has deployed the sign-language system, and who holds the model.
- Embodiment-SUDA (Hangzhou Sim2Real winner). Ask whether the policy transfers to a substation the team has never seen.
This radar runs weekly, on the results each window produces. The next issue covers 25 October to 1 November 2026.
References
- 1Infineon names winners of humanoid robotics startup challenge
roboticsandautomationnews.com
- 2Infineon Startup Challenge 2026: An Overview of the Winners
silicon-saxony.de
- 3
- 4AI for Insurance Challenge × Tasco
beyondpilot.ai
- 5Nine teams, next stop Demo Day
facebook.com
- 6
- 7
- 8
- 9The Build for Everyone Hackathon was a success
linkedin.com
- 10
- 11Sim2Real Challenge at the 2026 China Robot and AI Competition
baijiahao.baidu.com
- 12
- 13Cal Hacks 13.0
calhacks.io
- 14Metropolis
monad.xyz
- 15Vultr: Agent Rush Hackathon
lablab.ai
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.