Browse categories
Explore
Fiverr Pro
English
$
USD
I will program nvidia isaac sim and reinforcement learning in isaac lab
Jiten T
About this gig
Deploy GPU-Accelerated Simulations & Reinforcement Learning Pipelines in NVIDIA Isaac Sim**
Need state-of-the-art reinforcement learning (RL) or physics-accurate simulation for quadrupeds, bipeds, or complex manipulators? I configure NVIDIA Isaac Sim and Isaac Lab (formerly Orbit) environments to build high-fidelity training pipelines.
**What I Offer:**
* **Asset Import (URDF to USD):** Importing your CAD models/URDFs to OpenUSD, tuning physics materials, mass properties, and collision meshes.
* **Isaac Lab Gym Tasks:** Creating custom reinforcement learning environments, setting up action/observation spaces, and designing reward functions.
* **RL Policy Training:** Training stable policies using Stable-Baselines3, rsl_rl, or custom PPO algorithms with CUDA-accelerated parallel environments.
* **ROS2 Bridge & Sim-to-Real:** Setting up communication bridges to deploy trained neural network policies on real physical hardware.
**Deliverables:**
* Documented Python workspace.
* USD files & custom task scripts.
* Pre-trained model weights.
* Docker environment.
*Please contact me before ordering to discuss your robot's DoF and RL task objectives!
Get to know Jiten T
Jiten T
Robotics Engineer
- FromIndia
- Member sinceFeb 2026
- Avg. response time1 hour
- Last delivery2 months
Languages
English, Gujarati, Hindi
If your robot must work in the real world — not just in theory — I can help.
I’m a Robotics Engineer specializing in autonomous robots, mechanical design, and system integration. I’ve built real-world robots including surveillance, agricultural robots, and advanced robotic mechanisms.
From concept → design → working prototype, I focus on practical solutions that work outside the lab.
Other AI Development Services I Offer
FAQ
Q: Why do you recommend NVIDIA Isaac Sim over Gazebo for AI training?
A:** Isaac Sim is built on NVIDIA Omniverse and uses the PhysX physics engine, which runs directly on the GPU. This allows you to simulate thousands of robots in parallel at the same time, accelerating reinforcement learning training times from weeks to hours.
Q: How do we transfer the trained policy to a real robot?
A:** I set up the policy inference script in Python/C++ to read real-time joint states from your hardware (via ROS2 or serial communication), pass them through the trained model, and output target torques or velocities back to the motors.
Q: Do you design the reinforcement learning reward functions?
A:** Yes. I customize the reward functions based on your task. For example, for quadruped walking, I include rewards for target tracking velocity, body height stability, and foot contact, along with penalties for high torques, joint velocity spikes, and collisions.
Q: Can you run the training inside a headless server?
A:** Yes. I package the entire Isaac Sim workspace inside a Docker container configured to run headlessly on your GPU-enabled cloud instances (AWS, GCP) or local server, allowing you to train models over SSH.
