autona ROBOTICS
← ALL ARC SOFTWARE

ARC ENVIRONMENTS

Build the task.
Train the robot.

Task-specific simulation environments for robot reinforcement learning. Define the robot, scene, observations, actions, rewards, resets, and success criteria in a versioned environment that training teams can reuse.

Talk about Environments How it works

AUTONA ROBOT CLOUD / ROBOT RL ENVIRONMENTS

ARC / ENVIRONMENTSROBOT RL ENVIRONMENTS
ONE CONNECTED ROBOT CLOUDSYSTEM WORKFLOW
START WITH
A task brief, robot definition, and environment reference
WORK TOWARD
A versioned RL environment with resets, rewards, and evaluation cases
BUILT FOR
Robot-learning teams and research labs
INSIDE ARC ENVIRONMENTS

Design the environment around the task.

YAM pipe-flange fastening with scene context, observation/action contracts, rewards, resets, and held-out evaluation. Interface settings are illustrative; the scene is an authored simulation example.

ARC Environments interface preview. YAM pipe-flange fastening with scene context, observation/action contracts, rewards, resets, and held-out evaluation. Interface settings are illustrative; the scene is an authored simulation example.
EXAMPLE ENVIRONMENTYAM / PIPE-FLANGE FASTENING

Two arms.
One fastening task.

A supported pipe assembly, paired flanges, and two nuts already engaged on their studs. The recorded simulation covers grasp, turn, release, and regrasp.

Define observations and actions.

Workspace and wrist views, joint state, gripper state, and nut-to-stud alignment define the learning problem. The robot contract maps actions to the two arms.

Measure progress.

Track retained contact, engaged turns, axial travel, seating, and release. Build training variation around part poses, friction, and recovery cases.

Inspect the result.

Replay each turn, seating, and final withdrawal. Compare the recorded motion with the task’s completion criteria.

Watch the fastening simulation

Define the task.
Measure the learning.

Robot learning needs environments with useful observations, responsive physics, and measurable completion criteria. Connect a task reference to the scene, robot contract, rewards, and resets that define the training problem.

  1. Bring a task and a reference

    Start with the desired behavior, a video or scene description, a robot embodiment, and measurable completion criteria.

  2. Build and qualify the environment

    Model objects, fixtures, contacts, and control interfaces. Review scripted baselines and disclose idealized mechanics before using the environment for training.

  3. Train, vary, and evaluate

    Use a reproducible reset/step interface, track reward terms, and evaluate learned policies on held-out layouts and recovery cases.

Bring us a task for your robot.

Tell us about your robot, your team, and the problem you want to solve.

Start a conversation