SeReST Controller Fine-Tuning

This guide explains how to tune the SeReST (Safe Reparameterized Time) controller to track discrete paths (e.g., A* polylines) precisely, slow down smoothly near goals, and behave safely around dynamic obstacles.

It includes grouped parameters, recommended defaults, and practical “if you see X, change Y” advice.

Overview 

The SeReST controller is a real-time trajectory follower used in EasyNav. It blends path tracking, safety constraints, and curvature-aware motion limits into a single reactive control loop.

You will typically find it configured in:

controller_node:
  ros__parameters:
    controller_types: [serest]
    serest:
      plugin: easynav_serest_controller/SerestController
      rt_freq: 30.0

—

Quick-Start Defaults 

Start with these default parameters (adjust namespace as needed):

serest_controller:
  ros__parameters:
    # motion policy
    v_ref:                0.6
    max_linear_speed:     0.6
    max_angular_speed:    1.5
    max_linear_acc:       0.8
    max_angular_acc:      2.0
    allow_reverse:        false

    # tracking gains
    k_theta:              2.0
    k_y:                  1.2
    k_s:                  0.8
    ell:                  0.3

    # safety governor
    a_brake:              1.2
    a_lat_max:            1.5
    d0_margin:            0.20
    tau_latency:          0.10
    d_hard:               0.15
    t_emerg:              0.25

    # local heading smoothing
    blend_base:           0.6
    blend_k_per_v:        0.6
    kappa_max:            2.5

    # progress robustness
    v_progress_min:       0.05
    k_s_share_max:        0.5

    # goal behavior
    goal_pos_tol:         0.07
    goal_yaw_tol_deg:     6.0
    slow_radius:          0.60
    slow_min_speed:       0.03
    final_align_k:        2.0
    final_align_wmax:     0.6

    # corner guard (tight turns)
    corner_guard_enable:  true
    a_lat_soft:           1.1
    corner_gain_ey:       1.5
    corner_gain_eth:      0.7
    corner_gain_kappa:    0.4
    corner_min_alpha:     0.35
    corner_boost_omega:   0.8
    apex_ey_des:          0.05

—

Core Concepts 

Tracking controller: keeps the robot close to the reference path using heading and lateral-error feedback (k_theta, k_y, ell) plus a small longitudinal “suction” term (k_s).
Safety governor: dynamically re-parameterizes time, reducing forward progress when approaching obstacles or when curvature demands lower speeds.
Local heading smoothing: blends headings around corners to remove discontinuities while preserving the geometric path.
Goal behavior: decelerates within a configurable radius, aligning to the final yaw precisely.
Corner guard: limits lateral acceleration and boosts yaw authority to prevent drifting wide on tight turns.

—

Parameter-by-Parameter Guidance 

Tracking Gains 

``k_theta`` (1.5–3.0; default 2.0) Heading stiffness. ↑ if turns feel sluggish; ↓ if oscillating.
``k_y`` (0.8–2.0; default 1.2) Lateral correction authority. ↑ if robot cuts corners; ↓ if wobbly.
``ell`` (0.25–0.5 m; default 0.3) Lateral lookahead. ↑ to reduce twitchiness; ↓ for faster convergence.
``k_s`` (0.4–1.2; default 0.8) Longitudinal suction. ↑ if drifting along the segment; ↓ if resisting motion.

Motion Limits & Reference 

``v_ref`` (0.4–0.8 m/s) – Desired cruise speed.
``max_linear_speed`` / ``max_angular_speed`` – Respect hardware limits.
``max_linear_acc`` / ``max_angular_acc`` – ↑ if sluggish; ↓ if slipping.

Progress Robustness 

``v_progress_min`` (0.03–0.10) – Minimum forward motion; prevents stalls.
``k_s_share_max`` (0.3–0.7) – Caps suction influence; ↑ if stalling, ↓ if overshooting laterally.

Safety Governor 

``a_brake`` (1.0–2.0) – Deceleration limit for safe stopping.
``a_lat_max`` (1.2–2.5) – Max lateral acceleration for curvature limits.
``d0_margin`` (0.15–0.30) – Safety margin; ↑ for larger robots or noisy sensors.
``tau_latency`` (0.08–0.15 s) – Compensates sensor/control delay.
``d_hard`` / ``t_emerg`` – Emergency stop thresholds; ↑/↓ to adjust conservatism.

Local Heading Smoothing 

``blend_base`` (0.4–0.9) – Base length for corner blending.
``blend_k_per_v`` (0.4–0.9) – Extra smoothing proportional to velocity.
``kappa_max`` (2.0–3.5) – Curvature cap for feed-forward yawing.

Goal Behavior 

``slow_radius`` (0.5–1.0 m) – Begin slowing within this range.
``goal_pos_tol`` (0.05–0.10 m) – Position tolerance for success.
``goal_yaw_tol_deg`` (4–10°) – Final yaw tolerance.
``slow_min_speed`` (0.02–0.05) – Maintains slow progress.
``final_align_k`` / ``final_align_wmax`` – Tune final rotation precision.

Corner Guard 

``corner_guard_enable`` (bool) – Enables corner control logic.
``a_lat_soft`` (0.9–1.4) – Corner-specific accel cap; ↓ if drifting, ↑ if braking too much.
``corner_gain_ey`` (1.0–2.0) – Penalizes outside drift.
``corner_gain_eth`` (0.5–1.0) – Penalizes heading error in turns.
``corner_gain_kappa`` (0.3–0.7) – Penalizes curvature magnitude.
``corner_min_alpha`` (0.3–0.5) – Minimum speed scaling in corners.
``corner_boost_omega`` (0.6–1.0) – Yaw boost when outside curve.
``apex_ey_des`` (0.03–0.08) – Small inward bias to stay centered at apex.

—

Troubleshooting by Symptom 

Robot “moonwalks” (tries to go backward)

Set allow_reverse: false (default). If it still pulls backward, reduce k_s and increase v_progress_min.

Stalls near corners or from standstill 

Increase v_progress_min and/or k_s_share_max. Increase blend_base for smoother heading transitions.

Oscillates near goal 

Increase slow_radius or goal_pos_tol. Lower slow_min_speed; reduce final_align_k or final_align_wmax if overshooting.

Cuts corners or rides outside 

Enable corner guard. Decrease a_lat_soft. Increase corner_gain_ey and/or corner_boost_omega. If still wide, raise blend_base or reduce v_ref.

Overreacts laterally (“nervous”)

Increase ell; reduce k_y. If only in corners, lower corner_boost_omega.

Feels sluggish to turn 

Increase k_theta. Verify max_angular_speed and max_angular_acc are not too low. Reduce corner_gain_eth or corner_gain_kappa if over-limiting.

Emergency stops too often 

Reduce conservatism: increase t_emerg, reduce d_hard or d0_margin. Only do this if sensing latency is well-characterized.

Obstacle distance seems too pessimistic 

Lower lethal_cost_threshold (if available) or reduce dist_search_radius. Consider publishing measured closest_obstacle_distance in NavState.

—

Recommended Tuning Order 

Set kinematics: choose safe max_* and v_ref values.
Tune tracking: adjust k_theta, k_y, ell for stable path following.
Verify safety: confirm emergency stop triggers correctly.
Optimize corners: tune blend_*, a_lat_soft, and corner gains.
Smooth goal arrival: adjust slow_radius and final_align_*.
Ensure progress: tweak v_progress_min and k_s_share_max if stalling occurs.

—

FAQ 

Q: Can I enable reverse motion? A: Yes. Set allow_reverse: true. Tune goal and slow behavior carefully to prevent unwanted reversals.

Q: My path is jagged—what should I do? A: Increase blend_base and/or blend_k_per_v. Reduce kappa_max and v_ref slightly.

Q: How aggressive should ``a_lat_soft`` be? A: For indoor robots with good traction, 1.0–1.4 m/s² works well. On slippery or uneven surfaces, use 0.8–1.0 m/s² for safety.