# Intelligent Machines (robotics company)

> Source: https://aiwiki.ai/wiki/intelligent_machines_robotics
> Updated: 2026-10-27
> Fact-checked: 2026-10-27
> Categories: AI Companies, Embodied AI, Reinforcement Learning, Robotics Companies
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Intelligent Machines (robotics company)." aiwiki.ai, 27 Oct 2026. https://aiwiki.ai/wiki/intelligent_machines_robotics
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**Intelligent Machines** is a San Francisco robotics startup co-founded by Ashish Kumar, who previously led the AI team for [Tesla](https://aiwiki.ai/wiki/tesla)'s [Optimus](https://aiwiki.ai/wiki/tesla_optimus) [humanoid robot](https://aiwiki.ai/wiki/humanoid_robot) program. The Information reported on 7 October 2026, citing two people with direct knowledge, that the startup is developing non-humanoid robots to handle tasks normally done by highly skilled workers in industries such as manufacturing and aerospace.[1] According to Dealroom's summary of that report, the company uses [reinforcement learning](https://aiwiki.ai/wiki/reinforcement_learning) to teach robots complex tasks, is focused on software for now, has hired several former members of the Optimus team, and is raising a seed round that one person said could end up at around $100 million.[2] The coverage named no investors and no valuation, and it described the round as still being raised rather than closed.[2][3]

Public information about the company comes almost entirely from that report, which is paywalled, and from outlets that summarized it.[1][2][3]

## Overview

| Field | Detail |
|---|---|
| Name | Intelligent Machines |
| Headquarters | San Francisco, California[2][3] |
| Co-founder | Ashish Kumar, former Optimus AI lead at Tesla (other co-founders were not named in the coverage)[1][2] |
| Focus | Non-humanoid robots for skilled work in industries such as manufacturing and aerospace[1] |
| Technical approach | Reinforcement learning; software-first, with hardware design discussed by its leaders[2] |
| Staff | Includes several former members of Tesla's Optimus team[2][3] |
| Funding | Seed round being raised as of October 2026; reported to possibly reach around $100 million[2][3] |
| Investors | Not disclosed |
| Valuation | Not disclosed |

## Public disclosure

The company became public through an exclusive in The Information's AI Agenda newsletter, written by Kevin McLaughlin and published on 7 October 2026 under the headline "Former Tesla Optimus Lead is Launching an Industrial Robot Startup".[1] The free summary of the article says Kumar "has co-founded a startup", attributing the information to two people with direct knowledge, so the report did not come from a company announcement.[1] Dealroom's news digest, which summarized the story the same day, gave the company's name as Intelligent Machines and its location as San Francisco.[2] Crypto Briefing published an article on 8 October 2026 that also named the company.[3]

The founding date was not given in this coverage. Kumar left Tesla in September 2025,[4][6] and, according to Dealroom's summary of The Information's report, then spent about four months at [Meta](https://aiwiki.ai/wiki/meta_ai).[2]

## Technology and approach

The Information described the startup's robots as non-humanoid machines aimed at a range of tasks that would normally fall to highly skilled workers, with manufacturing and aerospace given as example industries.[1] Crypto Briefing described this as the opposite of a general-purpose humanoid such as Optimus.[3] It summarized the two positions this way: humanoid advocates argue that a human-shaped robot can work in factories designed for people without redesigning them, while specialists argue that most valuable industrial work is narrow enough for a custom machine to win on speed, reliability and cost.[3] For background on the specialized side, see [industrial robot](https://aiwiki.ai/wiki/industrial_robot).

According to Dealroom's summary, the company uses reinforcement learning, the subject of Kumar's doctoral work at [UC Berkeley](https://aiwiki.ai/wiki/uc_berkeley), to teach robots complex tasks.[2] It was focused on software at the time of the report, though its leaders had discussed designing hardware.[2] The report said it was not clear whether the company relies on [vision-language-action models](https://aiwiki.ai/wiki/vision_language_action_model), and it contrasted the company's approach with that of [world model](https://aiwiki.ai/wiki/world_model) startups such as [General Intuition](https://aiwiki.ai/wiki/general_intuition) and [World Labs](https://aiwiki.ai/wiki/world_labs).[2]

### Research lineage

Kumar's academic work centered on control policies trained with reinforcement learning in simulation and then deployed on real robots without further tuning, a problem usually called [sim-to-real transfer](https://aiwiki.ai/wiki/sim_to_real_transfer).[7] Several of these papers were co-authored with Jitendra Malik of UC Berkeley and [Deepak Pathak](https://aiwiki.ai/wiki/deepak_pathak) of Carnegie Mellon University.[11][12]

- **Rapid Motor Adaptation (RMA), 2021.** Kumar was first author of "RMA: Rapid Motor Adaptation for Legged Robots" with Zipeng Fu, Pathak and Malik, presented at RSS 2021.[8] RMA pairs a base policy with an adaptation module so that a legged robot can adapt to new conditions "in fractions of a second". It was trained completely in simulation, without reference trajectories or predefined foot trajectory generators, and deployed on the A1 [quadruped robot](https://aiwiki.ai/wiki/quadruped_robot) without fine-tuning.[8]
- **A-RMA, 2022.** "Adapting Rapid Motor Adaptation for Bipedal Robots" extended the method to the bipedal robot Cassie by fine-tuning the base policy with model-free reinforcement learning, and reported zero-shot deployment of a single policy to the real robot.[9]
- **In-hand rotation, 2022.** "In-Hand Object Rotation via Rapid Motor Adaptation", with Haozhi Qi as co-first author,[7] trained a fingertip controller entirely in simulation on cylindrical objects and deployed it without fine-tuning to a real [robot hand](https://aiwiki.ai/wiki/dexterous_hand), which then rotated dozens of objects of different sizes, shapes and weights.[10]
- **Vision-based locomotion, 2022.** "Legged Locomotion in Challenging Terrains using Egocentric Vision", with Ananye Agarwal as co-first author,[7] presented what the authors called the first end-to-end locomotion system able to traverse stairs, curbs, stepping stones and gaps, using a single front-facing depth camera on a medium-sized quadruped.[11] The project page lists it as winner of the Best Systems Paper Award at the Conference on Robot Learning (CoRL) 2022.[12]

When Kumar left Tesla, he summarized the Optimus AI team's direction on X: the team "went all-in on scalable methods", replacing the classical stack with reinforcement learning and "scaling dexterity by learning from videos".[6] Reinforcement learning is also the method that the reporting on Intelligent Machines attributes to the new company.[2][3]

## Founder

### Ashish Kumar

Kumar earned a PhD at the University of California, Berkeley; Electrek reported that the degree, in artificial intelligence, was awarded in 2023 and that he joined Tesla's humanoid effort soon afterward.[4][7] Earlier, he was first author of "Resource-efficient Machine Learning in 2 KB RAM for the Internet of Things" (ICML 2017, with Saurabh Goyal and Manik Varma), which introduced Bonsai, a tree-based algorithm for running predictions on microcontrollers such as the Arduino Uno.[13] His personal homepage lists invited talks at institutions including MIT, Stanford, the University of Pennsylvania and Facebook AI Research, among them a talk titled "Advancing Robot Intelligence with Reinforcement Learning" in Stanford's deep reinforcement learning course (CS-224R).[7]

At Tesla, Kumar led the Optimus AI team.[4][6] Sources differ on how long he held that role. Electrek wrote in September 2025 that he had led the team "for the past two years", while Dealroom's summary of The Information's report says he spent more than two years at Tesla, including a year as Optimus AI lead.[2][4] Electrek wrote that he was believed to be leading the AI side of the program, while Milan Kovac led the overall program, including robotics.[4]

In a post on X in September 2025, Kumar wrote that he had "Decided to leave Tesla", calling it "an incredible ride leading the Optimus AI team".[6] Electrek reported that he confirmed joining Meta's AI team as a researcher, and Drive Tesla Canada reported that his LinkedIn profile listed him as a research scientist at Meta.[4][5] When some Tesla shareholders accused him of leaving for money, he replied that the "Financial upside at Tesla was significantly larger" and that "If I wanted to optimize for money, I would have stayed at Tesla."[4] His departure followed Kovac's exit from Tesla in June 2025; Drive Tesla Canada reported that oversight of Optimus then rested with Ashok Elluswamy, Tesla's vice president of AI software.[5] According to The Information, as summarized by Dealroom, Kumar spent four months at Meta before starting the new company.[2]

### Selected publications

| Year | Title | Venue | Co-authors |
|---|---|---|---|
| 2017 | Resource-efficient Machine Learning in 2 KB RAM for the Internet of Things[13] | ICML 2017 | Saurabh Goyal, Manik Varma |
| 2021 | RMA: Rapid Motor Adaptation for Legged Robots[8] | RSS 2021 | Zipeng Fu, Deepak Pathak, Jitendra Malik |
| 2022 | Adapting Rapid Motor Adaptation for Bipedal Robots[9] | IROS 2022 (per author homepage)[7] | Zhongyu Li, Jun Zeng, Deepak Pathak, Koushil Sreenath, Jitendra Malik |
| 2022 | In-Hand Object Rotation via Rapid Motor Adaptation[10] | CoRL 2022 | Haozhi Qi, Roberto Calandra, Yi Ma, Jitendra Malik |
| 2022 | Legged Locomotion in Challenging Terrains using Egocentric Vision[11] | CoRL 2022, Best Systems Paper Award[12] | Ananye Agarwal, Jitendra Malik, Deepak Pathak |

## Funding

| Round | Reported | Amount | Status | Investors | Sources |
|---|---|---|---|---|---|
| Seed | October 2026 | Could end up around $100 million, according to one person | Being raised; no close reported | Not disclosed | [2][3] |

The only funding figure in the coverage is the roughly $100 million seed target, which The Information attributed to a single person.[2] Crypto Briefing commented that a round of that size "would be an unusually large opening check for a company at this stage".[3] The coverage did not report a valuation or name any investors.[2][3]

## Context and reception

Crypto Briefing placed Kumar's exit in a pattern of senior departures from the Optimus program, after Kovac's in June 2025, and listed three things to watch: whether the company closes its seed round at the reported size, whether more Optimus engineers follow Kumar, and whether Tesla shows progress on generalization in its own robots.[3] It also named risks for the specialized approach: each robot design serves a smaller market, new tasks may require new hardware, and reinforcement learning "can be data-hungry and slow to transfer from simulation to the factory floor".[3]

Dealroom's note on the report called Intelligent Machines a likely unprofiled company and grouped it with other startups founded by Tesla alumni, citing DensityAI, which it linked to Tesla's Dojo project.[2]

## See also

- [List of neolabs](https://aiwiki.ai/wiki/list_of_neolabs)
- [Neolab](https://aiwiki.ai/wiki/neolab)
- [Tesla Optimus](https://aiwiki.ai/wiki/tesla_optimus)
- [Physical AI](https://aiwiki.ai/wiki/physical_ai)
- [Robot learning](https://aiwiki.ai/wiki/robot_learning)
- [Embodied AI](https://aiwiki.ai/wiki/embodied_ai)

## References

1. Kevin McLaughlin, The Information. ["Former Tesla Optimus Lead is Launching an Industrial Robot Startup"](https://www.theinformation.com/newsletters/ai-agenda/former-tesla-optimus-lead-launching-industrial-robot-startup). 7 October 2026. (Paywalled; cited for the publicly visible summary.)
2. Dealroom News. ["Former Tesla Optimus Lead is Launching an Industrial Robot Startup"](https://dealroom.co/news/info-ztj5yx-former-tesla-optimus-lead-is-launching-an-industrial-robot-startup/) (summary of The Information's report). 7 October 2026.
3. Diego Almada Lopez, Crypto Briefing. ["Tesla's former Optimus AI lead bets on robots that don't look like people"](https://cryptobriefing.com/tesla-optimus-ai-lead-non-humanoid-robots/). 8 October 2026.
4. Fred Lambert, Electrek. ["Tesla loses another Optimus robot leader, and upsets TSLA stock gamblers"](https://electrek.co/2025/09/19/tesla-loses-another-optimus-robot-leader-upsets-tsla-stock-gamblers/). 19 September 2025.
5. Matt Wilson, Drive Tesla Canada. ["Tesla's Optimus AI Leader Departs for Meta"](https://driveteslacanada.ca/news/teslas-optimus-ai-leader-departs-for-meta/). 19 September 2025.
6. Ashish Kumar (@ashishkr9311). ["Decided to leave Tesla."](https://x.com/ashishkr9311/status/1968827611133427772) X. 19 September 2025 (UTC).
7. Ashish Kumar. ["Ashish Kumar" (personal homepage)](https://ashish-kmr.github.io/). Accessed 27 October 2026.
8. Ashish Kumar, Zipeng Fu, Deepak Pathak, Jitendra Malik. ["RMA: Rapid Motor Adaptation for Legged Robots"](https://arxiv.org/abs/2107.04034). arXiv:2107.04034. 8 July 2021.
9. Ashish Kumar, Zhongyu Li, Jun Zeng, Deepak Pathak, Koushil Sreenath, Jitendra Malik. ["Adapting Rapid Motor Adaptation for Bipedal Robots"](https://arxiv.org/abs/2205.15299). arXiv:2205.15299. 30 May 2022.
10. Haozhi Qi, Ashish Kumar, Roberto Calandra, Yi Ma, Jitendra Malik. ["In-Hand Object Rotation via Rapid Motor Adaptation"](https://arxiv.org/abs/2210.04887). arXiv:2210.04887. 10 October 2022.
11. Ananye Agarwal, Ashish Kumar, Jitendra Malik, Deepak Pathak. ["Legged Locomotion in Challenging Terrains using Egocentric Vision"](https://arxiv.org/abs/2211.07638). arXiv:2211.07638. 14 November 2022.
12. Agarwal, Kumar, Malik, Pathak. ["Legged Locomotion in Challenging Terrain using Egocentric Vision" (project page)](https://vision-locomotion.github.io/). Accessed 27 October 2026.
13. Ashish Kumar, Saurabh Goyal, Manik Varma. ["Resource-efficient Machine Learning in 2 KB RAM for the Internet of Things"](https://proceedings.mlr.press/v70/kumar17a.html). Proceedings of the 34th International Conference on Machine Learning, PMLR 70:1935-1944. 2017.