Esacode

Simulation Early access

Test thousands of conditions before the first real one

Run the same scene under thousands of conditions: friction, lighting, sensor noise, object poses, seeds. EsaSim spreads the runs across your GPUs and shows where the design holds and where it breaks, while a change still costs a rerun.

A grid of simulation frames from a bin-picking scene under different lighting and object poses

Ready to run

Start from a ready-made flow, keep full control

Simulations are blocks in a blueprint: run them as they are, change their settings, or open their code. Demos for every account are on the Projects page.

Early access

Landing sweep

Drone landings on a moving pad under wind, gusts and sensor noise, split into parallel tasks, with a 3D replay.

Early access

Warehouse fleets

A warehouse of mobile robots run live, with deliveries, orders, batteries and incidents; fleet-size and layout sweeps compare options in batches.

Early access

Rocket launch and landing

A booster flight at Earth scale in the 3D view: lift-off, staging, the entry burn and a landing on a drone ship, with cameras on board.

Early access

Soft-tissue anatomy

Organs from a CT scan that sag and sway under gravity as you turn the body, with measured stiffness per tissue.

Early access

Humanoid locomotion

A Unitree G1, Unitree H1 or Apptronik Apollo learns to walk, climb stairs and get up after a fall, thousands at once on one GPU with MuJoCo Warp, then walks a circuit live in 3D.

In development

Any robot from its description

Training for robots EsaSim has never seen, from their MJCF or URDF and a category: legged, arms, wheeled, flying.

Early access

Accelerator benchmark

Checks a worker end to end: driver, memory, throughput, on NVIDIA, AMD, Intel, Huawei Ascend or Tenstorrent.

Early access

Your own simulator

A Python block runs Isaac Sim, Isaac Lab, MuJoCo or your own code with live telemetry, on the worker's Python or in the Docker image you choose.

Sweeps

One job, many replicas

Set the number of replicas and EsaSim splits the parameter grid between them. Each replica gets its own GPU and log stream, and the job finishes when all of them do.

Job detail with four replicas, each showing its unit, progress bar and episode count

Blueprints

Watch the sweep as it runs

Build a sweep as a blueprint and its results land on one page: metrics, a chart filling in while tasks report, a table of every scenario and a 3D replay of the runs you want to look at.

  • Live telemetry from every task, every second or two
  • A landing flown live in 3D while the sweep runs, then replays with a timeline, runs to compare and a follow camera
  • A 3D scene block that gives each object its model (from the EsaSim library, or your own glTF, OBJ, STL and PLY) and its trajectory
  • Videos of any run rendered on your workers with Blender or the built-in renderer
About blueprints
View page of a landing sweep: metrics, a 3D replay of a drone landing, a chart of success rate by wind and a text summary

Sim-to-real

Take fewer surprises to real hardware

A sweep ranks conditions by how close they come to failure, so real test time goes to the few cases that deserve it. When something still goes wrong on hardware, it becomes a scenario in the set and is tested on every later change.

In development: bridge workers. Mark a local worker as connected to a real robot or board, run the trained policy on it, and watch the real robot in the same views as its digital twin, with limits, a watchdog and an emergency stop enforced by the bridge.

From simulation to real hardware
Side by side: a robot arm grasping in simulation and the same grasp on a real arm in the lab

Try a sweep on your own GPU

The free plan runs blueprints such as the landing sweep on your own workers.