DEV Community

DRIX10
DRIX10

Posted on Originally published at blogs.drix10.com

AI and Robotics Applications #251

#tech #ai #coding

🤖 Robotics - Challenges in Robotics Startups

Robotics is brutally hard. 99.99% of robotics startups die within the first few years of launch. Yet, every year, thousands of talented engineers and founders enter the field—often unaware of the technical, operational, and commercial risks that routinely push even well-funded startups to the brink of collapse.

Key Points:

  • Technical Risks: Robotics startups face significant technical challenges, including complex hardware and software integration, sensor calibration, and motion planning.

  • Operational Risks: Robotics startups must also contend with operational risks, such as supply chain disruptions, manufacturing quality control, and logistics management.

  • Commercial Risks: Commercial risks, including market competition, pricing, and revenue models, can also be significant.

🔗 Resources:

  • Original post
  • Original source
  • Robotics Startups
  • Challenges in robotics startups

🤖 AI - The Noise Problem in AI Research

60K+ is honestly insane. Yes, the noise problem is real, but I also find the number kind of amazing. It says a lot about how much AI has grown and how many people now want to work on it. The hard part is that the reviewing system clearly hasn’t scaled with that growth.

Key Points:

  • The Noise Problem: The noise problem in AI research refers to the issue of false positives and false negatives in the review process, which can lead to a significant amount of noise in the research community.

  • Scaling the Reviewing System: The reviewing system has not scaled with the growth of AI research, leading to a backlog of papers and a decrease in the quality of reviews.

  • Impact on Research: The noise problem and the lack of scaling in the reviewing system can have a significant impact on the quality and validity of AI research.

🔗 Resources:

  • Original post
  • Original source
  • AI Research
  • The noise problem in AI research

🤖 Robotics - Lack of Hardware Experience in Robotics Startups

Most of these people have never worked with real hardware or robots, and it shows. The funny thing is that if I do see a vibe-built product that I like, what's stopping me from vibe-building it myself?

Key Points:

  • Lack of Hardware Experience: Many robotics startups lack experience with real hardware or robots, which can lead to a lack of understanding of the technical challenges involved.

  • Vibe-Building: The term "vibe-building" refers to the process of building a product or system based on intuition or gut feeling, rather than a deep understanding of the underlying technology.

  • Impact on Product Quality: The lack of hardware experience and the reliance on vibe-building can have a significant impact on the quality and reliability of the final product.

🔗 Resources:

  • Original post
  • Original source
  • Robotics Startups
  • Lack of hardware experience in robotics startups

🤖 Robotics - Agentic Object-SLAM

Agentic Object-SLAM! Early experiments testing Astra’s 4D capabilities. Got it to reconstruct and track tabletop objects and hand pose, then loaded the scene into MuJoCo for a robot to copy the human action.

Key Points:

  • Agentic Object-SLAM: Agentic Object-SLAM is a type of SLAM (Simultaneous Localization and Mapping) that involves the use of a robot to interact with and manipulate objects in the environment.

  • Astra’s 4D Capabilities: Astra is a 4D computer vision system that can reconstruct and track objects in 3D space over time.

  • MuJoCo: MuJoCo is a physics engine that can be used to simulate complex physical systems, including robots and objects.

🔗 Resources:

  • Original post
  • Original source
  • Agentic Object-SLAM
  • Astra’s 4D capabilities

🤖 AI - Reasoning Over Constraints in VLM

Yes, I was asking myself this too. One of my favorite attempts: https:// arxiv.org/pdf/2507.16815 We can use the VLM to reason over constraints and condition the actor in latent space, but I think we need either a much higher-frequency approach or a way to decide when reasoning is needed.

Key Points:

  • Reasoning Over Constraints: Reasoning over constraints involves using a model to reason about the constraints and limitations of a system or environment.

  • VLM: VLM (Vision-Language Model) is a type of model that can be used to reason about visual and linguistic information.

  • Conditioning the Actor: Conditioning the actor involves using the VLM to condition the behavior of an actor or agent in a system.

🔗 Resources:

  • Original post
  • Original source
  • AI Research
  • Reasoning over constraints in VLM

🤖 Robotics - MOSS, the Open Source Litter Picking Rover

MOSS, the open source litter picking rover currently runs on a Jetson Orin Nano Super and a RealSense. Next, we’re exploring a more affordable build with a Raspberry Pi 5 and cheaper navigation sensors. Ordered both today: - Mighty Camera ($92) - ST’s VL53L9CX 3D ToF evaluation

Key Points:

  • MOSS: MOSS is an open source litter picking rover that can be used to clean up litter and debris in public spaces.

  • Jetson Orin Nano Super: The Jetson Orin Nano Super is a type of computer hardware that can be used to power the MOSS rover.

  • RealSense: RealSense is a type of sensor that can be used to provide depth and spatial information to the MOSS rover.

🔗 Resources:

  • Original post
  • Original source
  • MOSS
  • Open source litter picking rover

🤖 AI - REACT: A Fully Spiking State-Space Model

REACT: A Fully Spiking State-Space Model for Real-Time Event-Driven Temporal Perception Geoffroy Keime, Nicolas Cuperlier, Benoit R. Cottereau https:// arxiv.org/abs/2609.19204 [𝚌𝚜.𝚁𝙾 𝚌𝚜.𝙰𝙸]

Key Points:

  • REACT: REACT is a fully spiking state-space model that can be used for real-time event-driven temporal perception.

  • Fully Spiking: Fully spiking models are a type of model that can be used to simulate the behavior of spiking neurons.

  • State-Space Model: State-space models are a type of model that can be used to describe the behavior of a system over time.

🔗 Resources:

  • Original post
  • Original source
  • AI Research
  • REACT: A fully spiking state-space model

🤖 Robotics - Reality Check: AGIBOT’s A3 Ultra Humanoid

Reality Check: three workplaces, five evidence tests, zero disclosures. On 16 September AGIBOT released a commercial case video for its A3 Ultra humanoid, showing the machine at work in three settings — an auto dealership, a hotel, and a convenience store. The tasks on screen

Key Points:

  • AGIBOT’s A3 Ultra Humanoid: AGIBOT’s A3 Ultra Humanoid is a type of humanoid robot that can be used for a variety of tasks, including customer service and logistics.

  • Commercial Case Video: The commercial case video shows the A3 Ultra Humanoid in action, performing tasks in three different settings.

  • Evidence Tests: The evidence tests refer to the various tests and evaluations that were conducted to demonstrate the capabilities of the A3 Ultra Humanoid.

🔗 Resources:

  • Original post
  • Original source
  • AGIBOT’s A3 Ultra Humanoid
  • Reality check

🤖 Robotics - Reinforcement Learning for Active Perception

Reinforcement Learning for Active Perception in Autonomous Navigation. [GitHub & Paper ] Most robots navigate as if their cameras were nailed in place. But perception is not passive. Animals move their heads and eyes constantly to decide where to go next. Robots should do

Key Points:

  • Reinforcement Learning: Reinforcement learning is a type of machine learning that involves training an agent to take actions in an environment to maximize a reward.

  • Active Perception: Active perception involves the use of sensors and other devices to actively gather information about the environment.

  • Autonomous Navigation: Autonomous navigation involves the use of sensors and other devices to navigate through an environment without human intervention.

🔗 Resources:

  • Original post
  • Original source
  • Reinforcement Learning
  • Active perception in autonomous navigation

🤖 Robotics - BitRobot Bolts App Countdown

Last day to claim the Day 1 Badge for those who signed up during the BitRobot Bolts App countdown. Claim expires 11:59PM ET, Sept 21 (3:59 AM UTC on September 22).

Key Points:

  • BitRobot Bolts App: The BitRobot Bolts App is a type of app that can be used to connect with other robotics enthusiasts and stay up-to-date on the latest robotics news and developments.

  • Day 1 Badge: The Day 1 Badge is a type of badge that can be claimed by users who signed up for the BitRobot Bolts App on the first day it was available.

  • Claim Expiration: The claim expiration refers to the deadline for users to claim their Day 1 Badge.

🔗 Resources:

Read More & Connect

Interactive version: blogs.drix10.com

Written by Drishtant Ghosh (Drix10), a technical founder and engineer working across AI systems, developer infrastructure, and cybersecurity.

Top comments (0)