About me

Hello all !

My name is Malhar Karade. About 2 months ago, I graduated from the most competitive engineering colleges of India in the subject of Mechanical Engineering - Indian Institute of Technology Gandhinagar (IIT-GN). Robotics, for almost 3 years now has been a subject of my interest. It all started with an academic project in electronics - Arduino based mobile car. I've been in the college's robotics club & been it's secretary for a semester. In my tenure we participated in competitions and I hosted 2 intra-club projects too (drone and a robo-soccer competition). I've seen robotics on a subject level as well as a project level. Both of them demand very different sort of skills - one needs you to be good at writing codes or making PCBs, while the other one needs you to be a good team member, one who is able to talk to a person who knows 0 about your domain & trust me, they never know anything more than 0. A guy good at PCBs will almost never understand what ROS can do. Of course, vice versa !

Project background and progress

Almost about a year ago, me and a friend (also mechanical engineer and some expertise of web and app dev too) started to work on a robotics project - Autonomous Mobile Robots. We were presented with an opportunity of Government Funding ($2600/₹2.5 Lakh) to develop a prototype for the same. The purpose of the prototype is to demonstrate the concept of Autonomous Navigation using a camera feed.

These are 2 wheeled robots that are supposed to move around while carrying a substantial payload along with them. We've developed the navigation stack over NVIDIA's Jetson Orin Nano using the Isaac ROS framework. The algorithm uses a depth camera feed - Orbbec Gemini 335L - to navigate a known terrain. The algorithm generates a cost map that, in layman's terms, shows which area is clear to move in & which are will have chances of collision. Depending on the availability, the same algorithm generates velocity commands for the robot to move from one place to another. The robot has an hierarchical architecture - the Jetson Orin Nano processor generates the velocity command which are transferred to the STM32 microcontroller, which in turn makes sure that the motors rotate at the same velocity as requested by Jetson Orin Nano.

So far we've managed to make the Isaac ROS based navigation stack. It has been successfully deployed over a rudimentary chassis. The chassis consists of an 8mm acrylic sheet, to which all the components are either bolted or double taped. We've used this chassis to deploy our algorithm and achieved descent navigation results so far. Please refer to this drive link for visuals* : https://drive.google.com/drive/folders/11ong97sQ4tRIycR9ym9VrH9-IrbDjFpe?usp=sharing

*here you might see Arduino instead of STM32. We burnt our STM32 in some iterations.

What are this project's overall goals? How will you complete them ?

The end goal of this project is to get a fully functional 2 wheeled robot that is able to carry 80-100kgs of payload from one point to another. Since we have finalized the algorithm, we believe that fabrication of the body/chassis is the next step. The chassis will be designed and subjected to iterations. These iterations tell us where we need to improve the chassis. Although most of the iterations and testing will be carried out in simulations, some fine tuning can't be completed without actual fabrication. Thus, we aim to get the proper set of machineries and it's relevant raw material. Once we've finalized the design virtually - CAD Design - we'll subject it to actual fabrication & stress testing iterations.

How will this funding be used?

The funding will be majorly used to complete fabrication of the body. We need $8500 for machines and it's relevant accessories. $1500 for it's raw materials. $1000 will be used to buy tools which are essential for processing the chassis - cordless drills, electric sanders, etc. Rest of the funds are to be used for outsourcing the parts which can't be fabricated on a desktop machine.

Breakdown at a glance :

1. $7500 - Large format 3D printer (bigger print volume - ~80cm*80cm) and it's relevant accessories

2. $1000 - Desktop 3D printers (30cm*30cm prints) and it's relevant accessories

3. $1000 - Chemically processed spools (PLA+ mostly, but we'd like to buy a stock of ABS too)

4. $500 - Chemicals like isopropyl alcohol and ethanol to process the prints and make them shiny

5. $1000 - Power Tools (cordless drills, electric sander, etc.), hand tools (pliers & cutters, hammer, etc.), bench hardware (screws, nuts, bolts, extrusions, etc.), measurement tools (vernier calipers, tape, thread pitch gauge & screw sizer, set square and marking tools), painting tools (primer for plastic, spray paint and it's accessories etc.)

6. $ 4000 - Outsourcing designs, in case the dimensions of the Large format 3D printer mentioned above are not sufficient.

Total = $15,000

Who is on your team?

We are a team of 3 people working on this project. Me and a friend of mine (as mentioned above, both mechanical engineers from Indian Institute of Technology Gandhinagar (IITGN). This friend of mine, Anmol Kumar has web dev and app dev skills too. He's been responsible for the development of many websites & apps which are still used by IITGN ecosystem. And, just this week, a student with expertise in electronics and ROS from Indian Institute of Information Technology Dharwad (IIIT Dharwad), another competitive engineering college in India, has joined us in this endeavor. Together we make a mutually compensating team where we have skills from mechanical engineering, electronics, ROS, and software dev.

What are the most likely causes and outcomes if this project fails?

Having spent the past year building and validating the foundation, the question of whether the core concept works or not has substantially de-risked. The primary risk that we face is more towards the end of it being "stalled".

Causes : Mechanical fabrication of large size, especially using 3D printers take long durations to print - ~49-53 hours. This gives the manufacturing process a fixed latency. If stages like aesthetics finalization and total integration require more trial-and-error cycles than expected, the project risks stalling against its target milestones and target deadlines. Also, another reason for needing iterations would be failure of models due to excess payload. If stacked with more payload than it can take, the chassis will break in a literal physical sense. That'd need us to make it again. Thus, I can say that we face a very real fail point of failure due to feedback loops.

Outcome : If iteration cycles slow down, the outcome is a narrower or lower-tier deliverable rather than a complete failure. Even in a stalled state, the foundational system, codebase and documentation are added with the knowledge of fabrication iterations. Of course we'd have a physical chassis, but it might not be up to the mark of certain payload.

How do we plan to avoid failing : The only way to avoid it would be to make sure we don't engage in over engineering the item. From what I've read and felt so far in mechanical fabrication, there is nothing such as 100% perfect. Accuracy of 85% will do it to. This is basically saying, do only the needed number of iterations, don't try to make it so perfect that you run out of time, money and resources. One another way this can happen is if the vendor goes out of stock for certain raw materials. We need to make sure that we procure a good enough stock beforehand/in advance.

How much money have you raised in the last 12 months, and from where?

From government grants - $2600