feat: integrate Theseus S1 support (Follower, Leader, and Direct modes)

[vinland100]

Title

feat: integrate Theseus S1 support (Follower, Leader, and Direct modes)

Description

This PR integrates full support for the Theseus S1 robotic arm into the EVO-RL framework. It enables Theseus S1 users (single or dual-arm configurations) to seamlessly participate in the closed-loop reinforcement learning workflow, including data collection, training, and deployment.

Key Changes

• Robot Implementation: Added S1Follower and BiS1Follower classes compatible with LeRobot interfaces.

• Teleoperation Modes:

• Integrated S1Leader for standard teleoperation.

• Added a "Direct" mode for hybrid/composite teleoperation scenarios.

• Hardware Abstraction: Optimized for Theseus S1 V2 (USB serial) communication.

• Workflow Enhancements:

• Pre-calibrated integration: Disabled framework-level secondary calibration (require_calibration=false) as the S1 handles calibration at the SDK level.

Hardware Compatibility

• Model: Theseus S1 Robotic Arm.
• Communication: V2 version (USB serial). Note: V1 (CAN-based) is no longer supported.
• Setup: Support for both single-arm and dual-arm (Bi-S1) configurations.

Verified Functionality

• Hardware initialization and SDK communication.
• Expert data collection (lerobot_record).
• Policy teleoperation and inference (lerobot_teleoperate / lerobot_value_infer).
• Human-in-the-loop mechanism validation.

Documentation

An operation guide has been added to docs/source/YHRG-S1.md (and summarized in the PR description) covering:

1. Hardware configuration and mounting.
2. Expert data collection steps.
3. Training and fine-tuning (ACP) workflows.
4. Deployment and evaluation.

Related Issues

Partially implements requirements for embodied AI hardware diversification.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.