Jump to content

Connect SuperML | Leeroopedia MCP: Equip your AI agents with best practices, code verification, and debugging knowledge. Powered by Leeroo — building Organizational Superintelligence. Contact us at founders@leeroo.com.

Implementation:ARISE Initiative Robosuite Wrapper Base

From Leeroopedia
Knowledge Sources
Domains Robotics, Software Architecture, Simulation
Last Updated 2026-02-15 07:00 GMT

Overview

The Wrapper class is the base class for all environment wrappers in robosuite, providing transparent delegation of environment methods with support for data collection, logging, and observation transformation.

Description

The Wrapper class implements the decorator pattern for robosuite environments, allowing transparent interception and modification of environment interactions. By default, all methods delegate directly to the wrapped environment, making the wrapper invisible to calling code. Subclasses override specific methods to add functionality like data logging, observation normalization, or action transformation.

The class provides default implementations for the core environment interface: step, reset, render, observation_spec, action_spec, and action_dim. The unwrapped property recursively retrieves the innermost environment, bypassing all wrapper layers.

A safety mechanism against double-wrapping is provided via _warn_double_wrap, which traverses the wrapper chain and raises an exception if the same wrapper class appears more than once. This prevents accidental duplicate application of wrappers.

The __getattr__ fallback method ensures that any method or attribute not explicitly defined on the wrapper is transparently forwarded to the wrapped environment. For callable attributes, it wraps the return value to prevent the wrapped environment from "leaking" out - if a wrapped method returns the environment itself, the wrapper returns self instead, maintaining the wrapper chain.

Usage

Use this class as the base for custom environment wrappers. Subclass it and override the methods you need to modify. Common use cases include data collection wrappers, reward shaping wrappers, observation normalization wrappers, and gym-compatible interface wrappers.

Code Reference

Source Location

Signature

class Wrapper:
    def __init__(self, env)

    @classmethod
    def class_name(cls) -> str

    def _warn_double_wrap(self) -> None

    def step(self, action)

    def reset(self)

    def render(self, **kwargs)

    def observation_spec(self)

    @property
    def action_spec(self)

    @property
    def action_dim(self) -> int

    @property
    def unwrapped(self)

Import

from robosuite.wrappers.wrapper import Wrapper

I/O Contract

Inputs

Name Type Required Description
env MujocoEnv Yes The robosuite environment to wrap
action np.ndarray Yes (step) Action to take in the environment

Outputs

Name Type Description
step() tuple (observations, reward, done, info) from the wrapped environment
reset() OrderedDict Observation dictionary after environment reset
observation_spec() OrderedDict Observation space specification
action_spec tuple(np.ndarray, np.ndarray) (low, high) action limits
action_dim int Dimensionality of the action space
unwrapped MujocoEnv The innermost unwrapped environment
class_name() str The wrapper class name

Usage Examples

import numpy as np
from robosuite.wrappers.wrapper import Wrapper
import robosuite

# Basic wrapping of an environment
env = robosuite.make("Lift", robots="Panda")
wrapped_env = Wrapper(env)

# The wrapper delegates all calls transparently
obs = wrapped_env.reset()
obs, reward, done, info = wrapped_env.step(np.zeros(wrapped_env.action_dim))

# Access the unwrapped environment
original_env = wrapped_env.unwrapped

# Create a custom wrapper subclass
class LoggingWrapper(Wrapper):
    def __init__(self, env):
        super().__init__(env)
        self.step_count = 0

    def step(self, action):
        self.step_count += 1
        obs, reward, done, info = self.env.step(action)
        print(f"Step {self.step_count}: reward={reward:.4f}")
        return obs, reward, done, info

logged_env = LoggingWrapper(env)

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.
Principle
Implementation
Heuristic
Environment