evaluation.py

def evaluate_simulation_outcome(pybullet_feedback, obstacle_far_edge_x=0.8):
    """Evaluate the simulation outcome based on PyBullet feedback"""
    
    # Extract robot position
    robot_x_pos = pybullet_feedback['robot_position'][0]
    
    # Check if robot crossed the obstacle
    robot_crossed_obstacle = robot_x_pos > obstacle_far_edge_x
    
    # Check if robot remains upright
    robot_remains_upright = pybullet_feedback['is_robot_upright']
    
    # Check for significant collision during pass
    # Simplified logic: If robot crossed, minor contacts are okay
    # If robot didn't cross and had contacts, that's a failure
    obstacle_contacts_exist = pybullet_feedback['obstacle_contacts_exist']
    
    if robot_crossed_obstacle:
        # If robot crossed, we consider it successful regardless of minor contacts
        no_significant_collision = True
    else:
        # If robot didn't cross and had contacts, it's likely stuck
        no_significant_collision = not obstacle_contacts_exist
    
    # Overall success requires all criteria to be met
    overall_success = (robot_crossed_obstacle and 
                      robot_remains_upright and 
                      no_significant_collision)
    
    # Determine specific failure point
    specific_failure_point = "none"
    if not robot_remains_upright:
        specific_failure_point = "fell_over"
    elif obstacle_contacts_exist and not robot_crossed_obstacle:
        specific_failure_point = "collided_and_stuck"
    elif not robot_crossed_obstacle:
        specific_failure_point = "failed_to_reach_or_cross"
    
    evaluation_results = {
        "robot_crossed_obstacle": robot_crossed_obstacle,
        "no_significant_collision_with_obstacle_during_pass": no_significant_collision,
        "robot_remains_upright": robot_remains_upright,
        "overall_success": overall_success,
        "specific_failure_point": specific_failure_point,
        "final_robot_x_position": robot_x_pos
    }
    
    return evaluation_results

def evaluate_simulation_outcome_with_criteria(pybullet_feedback, obstacle_far_edge_x=0.8, llm_success_conditions=None):
    """Enhanced evaluation that maps simulation results to LLM-defined success criteria"""
    
    # First get the standard evaluation
    standard_eval = evaluate_simulation_outcome(pybullet_feedback, obstacle_far_edge_x)
    
    # Add criteria-specific evaluation
    if llm_success_conditions:
        criteria_evaluation = {}
        
        for condition in llm_success_conditions:
            condition_lower = condition.lower()
            
            # Map LLM conditions to simulation results
            if "cross" in condition_lower and "obstacle" in condition_lower:
                criteria_evaluation[condition] = "ACHIEVED" if standard_eval['robot_crossed_obstacle'] else "FAILED"
            
            elif "stability" in condition_lower or "stable" in condition_lower or "upright" in condition_lower:
                criteria_evaluation[condition] = "ACHIEVED" if standard_eval['robot_remains_upright'] else "FAILED"
            
            elif "collision" in condition_lower or "stuck" in condition_lower:
                criteria_evaluation[condition] = "ACHIEVED" if standard_eval['no_significant_collision_with_obstacle_during_pass'] else "FAILED"
            
            elif "reach" in condition_lower and "position" in condition_lower:
                criteria_evaluation[condition] = "ACHIEVED" if standard_eval['robot_crossed_obstacle'] else "FAILED"
            
            elif "quick" in condition_lower or "fast" in condition_lower or "efficient" in condition_lower:
                # For speed criteria, consider time and distance
                distance_ratio = standard_eval['final_robot_x_position'] / obstacle_far_edge_x if obstacle_far_edge_x > 0 else 0
                criteria_evaluation[condition] = "ACHIEVED" if distance_ratio > 0.9 else "PARTIALLY_ACHIEVED" if distance_ratio > 0.5 else "FAILED"
            
            elif "stop" in condition_lower or "halt" in condition_lower:
                # For stopping criteria, assume achieved if robot is stable and crossed
                criteria_evaluation[condition] = "ACHIEVED" if (standard_eval['robot_crossed_obstacle'] and standard_eval['robot_remains_upright']) else "FAILED"
            
            elif "land" in condition_lower:
                # For landing criteria (drones), assume achieved if stable and crossed
                criteria_evaluation[condition] = "ACHIEVED" if (standard_eval['robot_crossed_obstacle'] and standard_eval['robot_remains_upright']) else "FAILED"
            
            else:
                # Generic condition - map to overall success
                criteria_evaluation[condition] = "ACHIEVED" if standard_eval['overall_success'] else "UNDETERMINED"
        
        standard_eval["criteria_evaluation"] = criteria_evaluation
    
    return standard_eval

def format_feedback_for_llm(evaluation_results):
    """Format evaluation results into human-readable feedback for LLM"""
    
    success_status = "Succeeded" if evaluation_results['overall_success'] else "Failed"
    
    # Build collision note
    if evaluation_results['no_significant_collision_with_obstacle_during_pass']:
        collision_note = "No impeding contacts"
    else:
        collision_note = "Contacts detected with obstacle"
    
    # Build failure reason
    failure_reason = evaluation_results['specific_failure_point'] if not evaluation_results['overall_success'] else 'N/A'
    
    feedback = (
        f"Feedback: Obstacle passage attempt: {success_status}. "
        f"Crossed Obstacle: {evaluation_results['robot_crossed_obstacle']}. "
        f"Remained Upright: {evaluation_results['robot_remains_upright']}. "
        f"Collision Note: {collision_note}. "
        f"Failure reason: {failure_reason}. "
        f"Final X position: {evaluation_results['final_robot_x_position']:.2f}m."
    )
    
    return feedback

def format_feedback_for_llm_with_criteria(evaluation_results, llm_success_conditions=None):
    """Enhanced feedback format that includes LLM criteria evaluation"""
    
    # Start with standard feedback
    feedback = format_feedback_for_llm(evaluation_results)
    
    # Add criteria-specific feedback if available
    if llm_success_conditions and 'criteria_evaluation' in evaluation_results:
        feedback += "\n\nAssessment against your stated success conditions:"
        
        criteria_eval = evaluation_results['criteria_evaluation']
        for condition in llm_success_conditions:
            status = criteria_eval.get(condition, "UNDETERMINED")
            
            if status == "ACHIEVED":
                feedback += f"\n✅ '{condition}': ACHIEVED"
            elif status == "PARTIALLY_ACHIEVED":
                feedback += f"\n⚠️ '{condition}': PARTIALLY_ACHIEVED"
            elif status == "FAILED":
                feedback += f"\n❌ '{condition}': FAILED"
            else:
                feedback += f"\n❓ '{condition}': UNDETERMINED (based on core simulation checks)"
        
        # Add interpretation guidance
        feedback += f"\n\nCore simulation metrics:"
        feedback += f"\n- Distance traveled: {evaluation_results['final_robot_x_position']:.3f}m (target: >0.8m)"
        feedback += f"\n- Obstacle crossed: {'Yes' if evaluation_results['robot_crossed_obstacle'] else 'No'}"
        feedback += f"\n- Remained stable: {'Yes' if evaluation_results['robot_remains_upright'] else 'No'}"
        feedback += f"\n- Clean pass: {'Yes' if evaluation_results['no_significant_collision_with_obstacle_during_pass'] else 'No'}"
    
    return feedback