[Newton] Optimizes performance for some environments

[AntoineRichard]

Description

Perf opt to get the following:

Checklist

• I have read and understood the contribution guidelines
• I have run the pre-commit checks with ./isaaclab.sh --format
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• I have updated the changelog and the corresponding version in the extension's config/extension.toml file
• I have added my name to the CONTRIBUTORS.md or my name already exists there

[greptile-apps]

Greptile Overview

Greptile Summary

This PR implements performance optimizations for the Newton physics solver across multiple robot configurations and environment setups, yielding significant FPS improvements at scale (4096-16384 environments).

Key Changes:

• Solver tuning: Enabled ls_parallel=True and integrator="implicit" across direct environments (Ant, Cartpole, Humanoid); adjusted ls_iterations, njmax, and nconmax parameters per-environment
• Actuator friction removal: Removed small friction values (1e-5 to 0.0001) from most robot configs (Ant, Cassie, Franka, Humanoid, H1, G1, UR10) as a performance optimization
• API modernization: Refactored locomotion_env.py to use default_root_pose/default_root_vel instead of the deprecated default_root_state API
• A1 actuator fix: Re-enabled effort_limit_sim and friction parameters that were previously commented out
• Benchmark updates: Removed camera tasks and reduced environment counts (removed 1024, 2048) from scaling benchmarks

Confidence Score: 4/5

• This PR is safe to merge; changes are configuration-only with demonstrated performance benefits.
• All changes are configuration tuning (solver parameters, actuator friction values) and a minor API migration. The deprecated API usage was correctly updated. Performance improvements are supported by benchmark graphs in the PR description.
• The unitree.py file has the most significant actuator changes - verify A1 robot behavior with re-enabled friction and effort_limit_sim parameters.

Important Files Changed

File Analysis

Filename	Score	Overview
scripts/benchmarks/eval_all_for_newton_alpha_for_scaling.sh	4/5	Removed camera tasks and reduced env counts (1024, 2048 removed) for benchmarking; contains hardcoded path but existed before this PR.
source/isaaclab_assets/isaaclab_assets/robots/unitree.py	4/5	Major changes: re-enabled effort_limit_sim and friction for A1; removed friction from H1, G1, and G1_29_DOF configs.
source/isaaclab_tasks/isaaclab_tasks/direct/ant/ant_env.py	5/5	Reduced njmax/nconmax constraints, enabled ls_parallel=True and integrator="implicit" for better performance.
source/isaaclab_tasks/isaaclab_tasks/direct/cartpole/cartpole_env.py	5/5	Increased ls_iterations (3→10), added ls_parallel=True and integrator="implicit" for solver optimization.
source/isaaclab_tasks/isaaclab_tasks/direct/locomotion/locomotion_env.py	5/5	Refactored to use new API: default_root_pose/default_root_vel instead of deprecated default_root_state.

Sequence Diagram

sequenceDiagram
    participant Env as Environment
    participant Robot as Robot Articulation
    participant Solver as MJWarp Solver
    participant Data as Robot Data

    Note over Env,Solver: Performance-Optimized Reset Flow
    
    Env->>Robot: reset(env_ids)
    Robot->>Data: default_root_pose[env_ids]
    Robot->>Data: default_root_vel[env_ids]
    Data-->>Robot: pose (transform)
    Data-->>Robot: velocity (spatial_vector)
    
    Robot->>Robot: write_root_pose_to_sim()
    Robot->>Robot: write_root_velocity_to_sim()
    Robot->>Robot: write_joint_state_to_sim()
    
    Note over Solver: Solver Config Changes
    Solver->>Solver: ls_parallel=True
    Solver->>Solver: integrator="implicit"
    Solver->>Solver: Tuned njmax/nconmax/ls_iterations

Loading

[greptile-apps]

_{18 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[AntoineRichard]

Trainings is looking good against PhysX