Guillaume Bellegarda

Guillaume Bellegarda, Auke Ijspeert

In this letter, we present a method for integrating central pattern generators (CPGs), i.e. systems of coupled oscillators, into the deep reinforcement learning (DRL) framework to produce robust and omnidirectional quadruped locomotion. The agent learns to directly modulate the intrinsic oscillator setpoints (amplitude and frequency) and coordinate rhythmic behavior among different oscillators. This approach also allows the use of DRL to explore questions related to neuroscience, namely the role of descending pathways, interoscillator couplings, and sensory feedback in gait generation. We train our policies in simulation and perform a sim-to-real transfer to the Unitree A1 quadruped, where we observe robust behavior to disturbances unseen during training, most notably to a dynamically added 13.75 kg load representing 115% of the nominal quadruped mass. We test several different observation spaces based on proprioceptive sensing and show that our framework is deployable with no domain randomization and very little feedback, where along with the oscillator states, it is possible to provide only contact booleans in the observation space. Video results can be found at https://youtu.be/xqXHLzLsEV4.

Milad Shafiee, Guillaume Bellegarda, Auke Ijspeert

Learning a locomotion policy for quadruped robots has traditionally been constrained to a specific robot morphology, mass, and size. The learning process must usually be repeated for every new robot, where hyperparameters and reward function weights must be re-tuned to maximize performance for each new system. Alternatively, attempting to train a single policy to accommodate different robot sizes, while maintaining the same degrees of freedom (DoF) and morphology, requires either complex learning frameworks, or mass, inertia, and dimension randomization, which leads to prolonged training periods. In our study, we show that drawing inspiration from animal motor control allows us to effectively train a single locomotion policy capable of controlling a diverse range of quadruped robots. The robot differences encompass: a variable number of DoFs, (i.e. 12 or 16 joints), three distinct morphologies, a broad mass range spanning from 2 kg to 200 kg, and nominal standing heights ranging from 18 cm to 100 cm. Our policy modulates a representation of the Central Pattern Generator (CPG) in the spinal cord, effectively coordinating both frequencies and amplitudes of the CPG to produce rhythmic output (Rhythm Generation), which is then mapped to a Pattern Formation (PF) layer. Across different robots, the only varying component is the PF layer, which adjusts the scaling parameters for the stride height and length. Subsequently, we evaluate the sim-to-real transfer by testing the single policy on both the Unitree Go1 and A1 robots. Remarkably, we observe robust performance, even when adding a 15 kg load, equivalent to 125% of the A1 robot's nominal mass.

Guillaume Bellegarda, Milad Shafiee, Auke Ijspeert

We present a framework for learning visually-guided quadruped locomotion by integrating exteroceptive sensing and central pattern generators (CPGs), i.e. systems of coupled oscillators, into the deep reinforcement learning (DRL) framework. Through both exteroceptive and proprioceptive sensing, the agent learns to coordinate rhythmic behavior among different oscillators to track velocity commands, while at the same time override these commands to avoid collisions with the environment. We investigate several open robotics and neuroscience questions: 1) What is the role of explicit interoscillator couplings between oscillators, and can such coupling improve sim-to-real transfer for navigation robustness? 2) What are the effects of using a memory-enabled vs. a memory-free policy network with respect to robustness, energy-efficiency, and tracking performance in sim-to-real navigation tasks? 3) How do animals manage to tolerate high sensorimotor delays, yet still produce smooth and robust gaits? To answer these questions, we train our perceptive locomotion policies in simulation and perform sim-to-real transfers to the Unitree Go1 quadruped, where we observe robust navigation in a variety of scenarios. Our results show that the CPG, explicit interoscillator couplings, and memory-enabled policy representations are all beneficial for energy efficiency, robustness to noise and sensory delays of 90 ms, and tracking performance for successful sim-to-real transfer for navigation tasks. Video results can be found at https://youtu.be/wpsbSMzIwgM.

Wanming Yu, Chuanyu Yang, Christopher McGreavy et al.

Robot motor skills can be learned through deep reinforcement learning (DRL) by neural networks as state-action mappings. While the selection of state observations is crucial, there has been a lack of quantitative analysis to date. Here, we present a systematic saliency analysis that quantitatively evaluates the relative importance of different feedback states for motor skills learned through DRL. Our approach can identify the most essential feedback states for locomotion skills, including balance recovery, trotting, bounding, pacing and galloping. By using only key states including joint positions, gravity vector, base linear and angular velocities, we demonstrate that a simulated quadruped robot can achieve robust performance in various test scenarios across these distinct skills. The benchmarks using task performance metrics show that locomotion skills learned with key states can achieve comparable performance to those with all states, and the task performance or learning success rate will drop significantly if key states are missing. This work provides quantitative insights into the relationship between state observations and specific types of motor skills, serving as a guideline for robot motor learning. The proposed method is applicable to differentiable state-action mapping, such as neural network based control policies, enabling the learning of a wide range of motor skills with minimal sensing dependencies.

Milad Shafiee, Guillaume Bellegarda, Auke Ijspeert

Quadruped animal locomotion emerges from the interactions between the spinal central pattern generator (CPG), sensory feedback, and supraspinal drive signals from the brain. Computational models of CPGs have been widely used for investigating the spinal cord contribution to animal locomotion control in computational neuroscience and in bio-inspired robotics. However, the contribution of supraspinal drive to anticipatory behavior, i.e. motor behavior that involves planning ahead of time (e.g. of footstep placements), is not yet properly understood. In particular, it is not clear whether the brain modulates CPG activity and/or directly modulates muscle activity (hence bypassing the CPG) for accurate foot placements. In this paper, we investigate the interaction of supraspinal drive and a CPG in an anticipatory locomotion scenario that involves stepping over gaps. By employing deep reinforcement learning (DRL), we train a neural network policy that replicates the supraspinal drive behavior. This policy can either modulate the CPG dynamics, or directly change actuation signals to bypass the CPG dynamics. Our results indicate that the direct supraspinal contribution to the actuation signal is a key component for a high gap crossing success rate. However, the CPG dynamics in the spinal cord are beneficial for gait smoothness and energy efficiency. Moreover, our investigation shows that sensing the front feet distances to the gap is the most important and sufficient sensory information for learning gap crossing. Our results support the biological hypothesis that cats and horses mainly control the front legs for obstacle avoidance, and that hind limbs follow an internal memory based on the front limbs' information. Our method enables the quadruped robot to cross gaps of up to 20 cm (50% of body-length) without any explicit dynamics modeling or Model Predictive Control (MPC).

Milad Shafiee, Guillaume Bellegarda, Auke Ijspeert

Quadruped animals seamlessly transition between gaits as they change locomotion speeds. While the most widely accepted explanation for gait transitions is energy efficiency, there is no clear consensus on the determining factor, nor on the potential effects from terrain properties. In this article, we propose that viability, i.e. the avoidance of falls, represents an important criterion for gait transitions. We investigate the emergence of gait transitions through the interaction between supraspinal drive (brain), the central pattern generator in the spinal cord, the body, and exteroceptive sensing by leveraging deep reinforcement learning and robotics tools. Consistent with quadruped animal data, we show that the walk-trot gait transition for quadruped robots on flat terrain improves both viability and energy efficiency. Furthermore, we investigate the effects of discrete terrain (i.e. crossing successive gaps) on imposing gait transitions, and find the emergence of trot-pronk transitions to avoid non-viable states. Compared with other potential criteria such as peak forces and energy efficiency, viability is the only improved factor after gait transitions on both flat and discrete gap terrains, suggesting that viability could be a primary and universal objective of gait transitions, while other criteria are secondary objectives and/or a consequence of viability. Moreover, we deploy our learned controller in sim-to-real hardware experiments and demonstrate state-of-the-art quadruped agility in challenging scenarios, where the Unitree A1 quadruped autonomously transitions gaits between trot and pronk to cross consecutive gaps of up to 30 cm (83.3 % of the body-length) at over 1.3 m/s.

Peizhuo Li, Hongyi Li, Ge Sun et al.

Despite recent advances in learning-based controllers for legged robots, deployments in human-centric environments remain limited by safety concerns. Most of these approaches use position-based control, where policies output target joint angles that must be processed by a low-level controller (e.g., PD or impedance controllers) to compute joint torques. Although impressive results have been achieved in controlled real-world scenarios, these methods often struggle with compliance and adaptability when encountering environments or disturbances unseen during training, potentially resulting in extreme or unsafe behaviors. Inspired by how animals achieve smooth and adaptive movements by controlling muscle extension and contraction, torque-based policies offer a promising alternative by enabling precise and direct control of the actuators in torque space. In principle, this approach facilitates more effective interactions with the environment, resulting in safer and more adaptable behaviors. However, challenges such as a highly nonlinear state space and inefficient exploration during training have hindered their broader adoption. To address these limitations, we propose SATA, a bio-inspired framework that mimics key biomechanical principles and adaptive learning mechanisms observed in animal locomotion. Our approach effectively addresses the inherent challenges of learning torque-based policies by significantly improving early-stage exploration, leading to high-performance final policies. Remarkably, our method achieves zero-shot sim-to-real transfer. Our experimental results indicate that SATA demonstrates remarkable compliance and safety, even in challenging environments such as soft/slippery terrain or narrow passages, and under significant external disturbances, highlighting its potential for practical deployments in human-centric and safety-critical scenarios.

Guillaume Bellegarda, Yiyu Chen, Zhuochen Liu et al.

Deep reinforcement learning has emerged as a popular and powerful way to develop locomotion controllers for quadruped robots. Common approaches have largely focused on learning actions directly in joint space, or learning to modify and offset foot positions produced by trajectory generators. Both approaches typically require careful reward shaping and training for millions of time steps, and with trajectory generators introduce human bias into the resulting control policies. In this paper, we present a learning framework that leads to the natural emergence of fast and robust bounding policies for quadruped robots. The agent both selects and controls actions directly in task space to track desired velocity commands subject to environmental noise including model uncertainty and rough terrain. We observe that this framework improves sample efficiency, necessitates little reward shaping, leads to the emergence of natural gaits such as galloping and bounding, and eases the sim-to-real transfer at running speeds. Policies can be learned in only a few million time steps, even for challenging tasks of running over rough terrain with loads of over 100% of the nominal quadruped mass. Training occurs in PyBullet, and we perform a sim-to-sim transfer to Gazebo and sim-to-real transfer to the Unitree A1 hardware. For sim-to-sim, our results show the quadruped is able to run at over 4 m/s without a load, and 3.5 m/s with a 10 kg load, which is over 83% of the nominal quadruped mass. For sim-to-real, the Unitree A1 is able to bound at 2 m/s with a 5 kg load, representing 42% of the nominal quadruped mass.

Guillaume Bellegarda, Chuong Nguyen, Quan Nguyen

In this paper, we consider a general task of jumping varying distances and heights for a quadrupedal robot in noisy environments, such as off of uneven terrain and with variable robot dynamics parameters. To accurately jump in such conditions, we propose a framework using deep reinforcement learning that leverages and augments the complex solution of nonlinear trajectory optimization for quadrupedal jumping. While the standalone optimization limits jumping to take-off from flat ground and requires accurate assumptions of robot dynamics, our proposed approach improves the robustness to allow jumping off of significantly uneven terrain with variable robot dynamical parameters and environmental conditions. Compared with walking and running, the realization of aggressive jumping on hardware necessitates accounting for the motors' torque-speed relationship as well as the robot's total power limits. By incorporating these constraints into our learning framework, we successfully deploy our policy sim-to-real without further tuning, fully exploiting the available onboard power supply and motors. We demonstrate robustness to environment noise of foot disturbances of up to 6 cm in height, or 33% of the robot's nominal standing height, while jumping 2x the body length in distance.

Guillaume Bellegarda, Katie Byl

Recent breakthroughs both in reinforcement learning and trajectory optimization have made significant advances towards real world robotic system deployment. Reinforcement learning (RL) can be applied to many problems without needing any modeling or intuition about the system, at the cost of high sample complexity and the inability to prove any metrics about the learned policies. Trajectory optimization (TO) on the other hand allows for stability and robustness analyses on generated motions and trajectories, but is only as good as the often over-simplified derived model, and may have prohibitively expensive computation times for real-time control. This paper seeks to combine the benefits from these two areas while mitigating their drawbacks by (1) decreasing RL sample complexity by using existing knowledge of the problem with optimal control, and (2) providing an upper bound estimate on the time-to-arrival of the combined learned-optimized policy, allowing online policy deployment at any point in the training process by using the TO as a worst-case scenario action. This method is evaluated for a car model, with applicability to any mobile robotic system. A video showing policy execution comparisons can be found at https://youtu.be/mv2xw83NyWU .

Guillaume Bellegarda, Katie Byl

Recent breakthroughs in the reinforcement learning (RL) community have made significant advances towards learning and deploying policies on real world robotic systems. However, even with the current state-of-the-art algorithms and computational resources, these algorithms are still plagued with high sample complexity, and thus long training times, especially for high degree of freedom (DOF) systems. There are also concerns arising from lack of perceived stability or robustness guarantees from emerging policies. This paper aims at mitigating these drawbacks by: (1) modeling a complex, high DOF system with a representative simple one, (2) making explicit use of forward and inverse kinematics without forcing the RL algorithm to "learn" them on its own, and (3) learning locomotion policies in Cartesian space instead of joint space. In this paper these methods are applied to JPL's Robosimian, but can be readily used on any system with a base and end effector(s). These locomotion policies can be produced in just a few minutes, trained on a single laptop. We compare the robustness of the resulting learned policies to those of other control methods. An accompanying video for this paper can be found at https://youtu.be/xDxxSw5ahnc .

Katherine Driggs-Campbell, Guillaume Bellegarda, Victor Shia et al.

This report describes a new experimental setup for human-in-the-loop simulations. A force feedback simulator with four axis motion has been setup for real-time driving experiments. The simulator will move to simulate the forces a driver feels while driving, which allows for a realistic experience for the driver. This setup allows for flexibility and control for the researcher in a realistic simulation environment. Experiments concerning driver distraction can also be carried out safely in this test bed, in addition to multi-agent experiments. All necessary code to run the simulator, the additional sensors, and the basic processing is available for use.